* Studio diffusion: cross-platform device policy, fp16 guard, lock split, validate-before-evict
Phase 1 of porting the richer diffusion stack onto the image-generation backend.
- Add a compartmentalized device/dtype policy module (diffusion_device.py)
resolving CUDA/ROCm/XPU/MPS/CPU with capability flags. Keeps the NVIDIA
capability-based bf16 choice; ROCm and XPU are isolated; MPS uses bf16 or
fp32, never a silent fp16 that renders a black image.
- Add a per-family fp16_incompatible flag (Z-Image) and promote a resolved
float16 to float32 for those families so they do not produce black images.
- Split the backend locks: a generation holds only _generate_lock, so status,
unload, and a new load are never blocked by a long denoise. Add per-generation
cancellation via callback_on_step_end so an eviction or a superseding load
preempts a running generation; a replacement load waits for it to stop before
allocating, so two pipelines never sit in VRAM at once.
- Validate a load request before the GPU handoff so an unloadable pick never
evicts a working chat model, and reject missing local paths up front.
- Add CPU-only tests for the device policy, dtype guard, lock split and
cancellation, and validate-before-evict, plus a GPU benchmark/regression
script (scripts/diffusion_bench.py) measuring latency, peak VRAM, and PSNR
against a saved reference.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Studio diffusion (Phase 2A): measured-budget memory planner + offload/VAE policy
Add a lean, backend-agnostic memory policy that picks a CPU-offload policy and
VAE tiling/slicing from measured free device memory vs the model's estimated
resident footprint, then applies it to the built pipeline. auto stays resident
when the model fits (byte-identical to the prior resident path), and falls to
whole-module offload when tight; fast/balanced/low_vram are explicit overrides.
Sequential submodule offload is unreliable for GGUF transformers on diffusers
0.38, so it falls back to whole-module offload and status reports the policy
actually engaged.
Verified on Z-Image-Turbo Q4_K_M (B200): auto reproduces the resident image with
no VRAM/latency regression (PSNR inf); balanced/low_vram cut generation peak VRAM
47.9% (15951 -> 8318 MB) with byte-identical output, at the expected latency cost.
73 prior + 35 new CPU tests pass.
* Studio diffusion (Phase 2D): streamed block-level offload + functional VAE tiling
Add a streamed 'group' offload tier (diffusers apply_group_offloading, block_level,
use_stream) that keeps the transformer flowing through the GPU a few blocks at a
time while the text encoder / VAE stay resident, and fix VAE tiling to drive the
VAE submodule (pipelines like Z-Image expose enable_tiling on pipe.vae, not the
pipeline). apply_memory_plan now returns the (policy, tiling) actually engaged so
status never overstates either, and group falls back to whole-module offload when
the transformer can't be streamed.
Measured on Z-Image (B200), all lossless (PSNR inf vs resident): balanced/group
cuts generation peak VRAM 32% (15951 -> 10840 MB) at near-resident speed (2.07 ->
2.99s); low_vram/model cuts it 48% (-> 8318 MB) but is slower (7.99s). Mode names
now match that tradeoff: balanced = stream the transformer, low_vram = offload
every component. auto picks group when the companions fit resident, else model.
112 CPU tests pass.
* Studio diffusion (Phase 5): image quality-vs-quant accuracy harness
Add scripts/diffusion_quality.py, the accuracy analogue of the KLD workflow: hold
prompt + seed fixed, render a grid with a reference quant (default BF16), then render
each candidate quant and measure drift from the reference. Records mean PSNR + SSIM
(pure-numpy, no skimage/scipy) and optional CLIP text-alignment + image-similarity
(transformers, --clip), plus file size, latency, and peak VRAM, then prints a
quality-vs-cost table and recommends the smallest quant within a quality budget.
--selftest validates the metrics on synthetic images with no GPU or model.
Verified on Z-Image (B200): the table degrades monotonically with quant size
(Q8 -> Q4 -> Q2: PSNR 21.7 -> 15.5, SSIM 0.82 -> 0.61), while CLIP-text stays flat
(~0.34) -- quantization erodes fine detail far more than prompt adherence.
* Studio diffusion (Phase 3): opt-in speed layer (channels_last / compile / TF32)
Add a speed_mode knob (off by default, so the render path stays bit-identical):
default applies channels_last VAE + regional torch.compile of the denoiser's
repeated block where eligible; max also enables TF32 matmul and fused QKV. Regional
compile is gated off for the GGUF transformer (dequantises per-op) and for families
flagged not compile-friendly (a new supports_torch_compile flag, False for Z-Image),
so it activates automatically only once a non-GGUF bf16 transformer is loaded. Speed
optims run before placement/offload, per the diffusers composition order. status now
reports speed_mode + the optims actually engaged.
Verified on Z-Image (B200): default -> ['channels_last'], max -> ['channels_last',
'tf32'], compile correctly skipped for GGUF; generation works in every mode.
121 CPU tests pass.
* Studio diffusion (Phase 2B): opt-in fp8 text-encoder layerwise casting
Add a text_encoder_fp8 knob that casts the companion text encoder(s) to fp8 (e4m3)
storage via diffusers apply_layerwise_casting, upcasting per layer to the bf16
compute dtype while normalisations and embeddings stay full precision. Applied
before placement, gated to CUDA + bf16, best-effort (a failure leaves the encoder
dense). status reports which encoders were cast.
Verified on Z-Image (B200, balanced/group mode where the encoder stays resident):
generation peak VRAM dropped 37% (10840 -> 6791 MB, below the lowest-VRAM offload)
at near-resident speed. It is a memory-vs-quality tradeoff, not free -- ~20 dB PSNR
vs the bf16 encoder, a larger shift than one transformer quant step -- so it is off
by default and documented as such, with the Phase 5 harness to size the cost.
127 CPU tests pass.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Studio diffusion (Phase 2C): NVFP4 text-encoder quant (+ generalise fp8 knob)
Generalise the text-encoder precision knob from a fp8 bool to text_encoder_quant
(fp8 | nvfp4). nvfp4 quantises the companion text encoder to 4-bit via torchao
NVFP4 weight-only (two-level microscaling) on Blackwell's FP4 tensor cores; fp8
stays the broader-hardware path (cc>=8.9). Both are gated, best-effort, and run
before placement; status reports the mode actually engaged. This is the lean
realisation of GGUF-native text-encoder quant: 4-bit on the encoder without the
3045-line port.
Verified on Z-Image (B200, balanced/group where the encoder stays resident), vs the
bf16 encoder: nvfp4 cut generation peak VRAM 48% (10840 -> 5593 MB, the lowest TE
option, below whole-model offload) at near-fp8 quality (16.4 vs 17.1 dB PSNR), and
both quants ran faster than bf16. A memory-vs-quality tradeoff (off by default);
size it per model with the Phase 5 quality harness. diffusion_bench gains
--text-encoder-quant.
129 CPU tests pass.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Studio diffusion (Phase 4): native stable-diffusion.cpp engine for CPU/Mac
Adds the CPU / Apple-Silicon tier of the two-engine strategy, mirroring the
chat backend's llama.cpp shell-out. Diffusers stays the default on CUDA / ROCm
/ XPU; this covers the hardware diffusers serves poorly, consuming the same
split GGUF assets Studio already curates.
- sd_cpp_args.py: pure sd-cli command builder. Maps the family to its
text-encoder flag (Z-Image Qwen3 to --llm, Qwen-Image to --qwen2vl, FLUX.1
CLIP-L + T5), and the diffusers memory policy (none/group/model/sequential)
to sd.cpp's offload flags (--offload-to-cpu / --clip-on-cpu / --vae-on-cpu /
--vae-tiling / --diffusion-fa), so one user knob drives both engines.
- sd_cpp_engine.py: SdCppEngine over a located sd-cli. find_sd_cpp_binary()
with the same precedence as the llama finder (env override, then the Studio
install root, then in-tree, then PATH), an is_available/version probe, and a
one-shot subprocess generate that streams progress and returns the PNG.
runtime_env() prepends the binary's directory to the platform library path
so a prebuilt's bundled libstable-diffusion.so resolves.
select_diffusion_engine() is the pure routing decision (GPU backends to
diffusers, CPU/MPS to native when present).
- install_sd_cpp_prebuilt.py: resolve + download the per-host prebuilt
(macOS-arm64/Metal, Linux x86_64 CPU, Vulkan/ROCm/Windows variants) into the
Studio install root. resolve_release_asset() is a pure, unit-tested
host-to-asset matrix.
- scripts/sd_cpp_smoke.py: end-to-end native generation harness.
Tests (CPU-only, subprocess/filesystem stubbed): 49 new across args, engine,
routing, runtime env, and the installer resolver. Full diffusion suite 166
passing.
Verified on a B200 box: built sd-cli (CUDA) and the prebuilt (CPU) both
generate Z-Image-Turbo Q4_K end to end through SdCppEngine: balanced (group
offload, 5.0s gen), low_vram (full CPU offload + VAE tiling, 13.4s), and the
dynamically-linked CPU prebuilt (50.4s on CPU), all producing coherent images.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Studio diffusion (Phase 6): img2img / inpaint / edit / LoRA / upscale on the native engine
Builds on Phase 4's native stable-diffusion.cpp engine, extending it from
text-to-image to the wider feature surface, since sd.cpp supports all of these
through the binary already. Pure command-builder additions plus one engine
method, so the txt2img path is unchanged.
- sd_cpp_args.py: SdCppGenParams gains image-conditioning fields. init_img +
strength make a run img2img, adding mask makes it inpaint, ref_images drives
FLUX-Kontext / Qwen-Image-Edit style editing (repeated --ref-image), and
lora_dir + the <lora:name:weight> prompt syntax select LoRAs. New
SdCppUpscaleParams + build_sd_cpp_upscale_command for the ESRGAN upscale run
mode (input image + esrgan model, no prompt / text encoders).
- sd_cpp_engine.py: the subprocess runner is factored into a shared _run() so
generate() (now carrying the conditioning flags) and a new upscale() reuse
the same streaming / error / output-check path.
- scripts/sd_cpp_smoke.py: --task {txt2img,img2img,upscale} with --init-img /
--strength / --upscale-model / --upscale-repeats.
Tests: 10 new across the img2img / inpaint / edit / LoRA flag construction, the
upscale builder and its validation, and the engine's img2img + upscale paths.
Full diffusion suite 176 passing.
Verified on a B200 box through SdCppEngine: img2img (Z-Image-Turbo Q4_K, the
init image conditioned at strength 0.6, 4.8s) and ESRGAN upscale
(512x512 -> 2048x2048 via RealESRGAN_x4plus_anime_6B, 2.7s), both producing
coherent images. Video and the diffusers-path feature wiring are deferred.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Studio diffusion (Phase 4): enforce the sd-cli timeout while reading output
Iterating proc.stdout directly blocks until the stream closes, so a sd-cli that hangs
without producing output (or without closing stdout) would never reach proc.wait and the
wall-clock timeout was silently bypassed. Drain stdout on a daemon thread and wait on the
PROCESS, so the main thread always enforces the timeout and kills a hung process (which
closes the pipe and ends the reader). Add a test that times out even when stdout blocks,
and make the no-binary test hermetic so a host-installed sd-cli can't leak in.
* Studio diffusion (Phase 4) review fixes: sd.cpp installer + engine hardening
- install_sd_cpp_prebuilt: download the release archive with urlopen + an explicit
timeout + copyfileobj (urlretrieve has no timeout and hangs on a stalled socket);
extract through a per-member containment check (Zip-Slip guard); expanduser the
--install-dir so a tilde path is not taken literally; and on Windows CUDA also fetch
the separately-published cudart runtime DLL archive so sd-cli.exe can start.
- sd_cpp_engine: find_sd_cpp_binary honors UNSLOTH_STUDIO_HOME / STUDIO_HOME like the
installer, so a custom-root install is discovered without UNSLOTH_SD_CPP_PATH; start
sd-cli with the parent-death child_popen_kwargs so it is not orphaned on a backend
crash; reap the SIGKILLed child (proc.wait) so a cancel/timeout does not leave a zombie.
- tests: Zip-Slip rejection, normal extraction, studio-home discovery.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Studio diffusion (Phase 4) review round 2: collect sd-cli batch outputs
Codex review: when batch_count > 1, stable-diffusion.cpp's save_results() writes
the numbered files <stem>_<idx><suffix> (base_0.png, base_1.png, ...) instead of
the literal --output path. SdCppEngine.generate checked only the literal path, so
a batch generation would exit 0 and then raise 'no image' (or return a stale
file). generate now returns the literal path when present and otherwise falls
back to the numbered siblings; single-image behavior is unchanged.
Test: a fake sd-cli that writes img_0.png/img_1.png (not img.png) is collected
without error.
* Studio diffusion (Phase 6) review round 2: img2img source dims + upscale repeats
Codex review on the native engine arg builder:
- build_sd_cpp_command emitted --width/--height unconditionally, so an
img2img/inpaint/edit run that left dims unset forced a 1024x1024 resize/crop of
the input. width/height are now Optional (None = unset): an image-conditioned
run (init_img or ref_images) with unset dims omits the flags so sd.cpp derives
the size from the input image (set_width_and_height_if_unset); a plain txt2img
run with unset dims keeps the prior 1024x1024 default; explicit dims are always
honored. width/height are read only by the builder, so the type change is local.
- build_sd_cpp_upscale_command used a truthiness guard (params.repeats and ...)
that silently swallowed repeats=0 into sd-cli's default of one pass, turning an
explicit no-op into a real upscale. It now rejects repeats < 1 with ValueError
and emits the flag for any explicit value != 1.
Tests: img2img unset dims omit width/height (init_img and ref_images), explicit
dims emitted, txt2img keeps 1024; upscale rejects repeats=0 and omits the flag at
the default. (Two pre-existing binary-discovery tests fail only because a real
sd-cli is installed in this dev environment; unrelated to this change.)
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: oobabooga <112222186+oobabooga@users.noreply.github.com>
* Studio diffusion: cross-platform device policy, fp16 guard, lock split, validate-before-evict
Phase 1 of porting the richer diffusion stack onto the image-generation backend.
- Add a compartmentalized device/dtype policy module (diffusion_device.py)
resolving CUDA/ROCm/XPU/MPS/CPU with capability flags. Keeps the NVIDIA
capability-based bf16 choice; ROCm and XPU are isolated; MPS uses bf16 or
fp32, never a silent fp16 that renders a black image.
- Add a per-family fp16_incompatible flag (Z-Image) and promote a resolved
float16 to float32 for those families so they do not produce black images.
- Split the backend locks: a generation holds only _generate_lock, so status,
unload, and a new load are never blocked by a long denoise. Add per-generation
cancellation via callback_on_step_end so an eviction or a superseding load
preempts a running generation; a replacement load waits for it to stop before
allocating, so two pipelines never sit in VRAM at once.
- Validate a load request before the GPU handoff so an unloadable pick never
evicts a working chat model, and reject missing local paths up front.
- Add CPU-only tests for the device policy, dtype guard, lock split and
cancellation, and validate-before-evict, plus a GPU benchmark/regression
script (scripts/diffusion_bench.py) measuring latency, peak VRAM, and PSNR
against a saved reference.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Studio diffusion (Phase 2A): measured-budget memory planner + offload/VAE policy
Add a lean, backend-agnostic memory policy that picks a CPU-offload policy and
VAE tiling/slicing from measured free device memory vs the model's estimated
resident footprint, then applies it to the built pipeline. auto stays resident
when the model fits (byte-identical to the prior resident path), and falls to
whole-module offload when tight; fast/balanced/low_vram are explicit overrides.
Sequential submodule offload is unreliable for GGUF transformers on diffusers
0.38, so it falls back to whole-module offload and status reports the policy
actually engaged.
Verified on Z-Image-Turbo Q4_K_M (B200): auto reproduces the resident image with
no VRAM/latency regression (PSNR inf); balanced/low_vram cut generation peak VRAM
47.9% (15951 -> 8318 MB) with byte-identical output, at the expected latency cost.
73 prior + 35 new CPU tests pass.
* Studio diffusion (Phase 2D): streamed block-level offload + functional VAE tiling
Add a streamed 'group' offload tier (diffusers apply_group_offloading, block_level,
use_stream) that keeps the transformer flowing through the GPU a few blocks at a
time while the text encoder / VAE stay resident, and fix VAE tiling to drive the
VAE submodule (pipelines like Z-Image expose enable_tiling on pipe.vae, not the
pipeline). apply_memory_plan now returns the (policy, tiling) actually engaged so
status never overstates either, and group falls back to whole-module offload when
the transformer can't be streamed.
Measured on Z-Image (B200), all lossless (PSNR inf vs resident): balanced/group
cuts generation peak VRAM 32% (15951 -> 10840 MB) at near-resident speed (2.07 ->
2.99s); low_vram/model cuts it 48% (-> 8318 MB) but is slower (7.99s). Mode names
now match that tradeoff: balanced = stream the transformer, low_vram = offload
every component. auto picks group when the companions fit resident, else model.
112 CPU tests pass.
* Studio diffusion (Phase 5): image quality-vs-quant accuracy harness
Add scripts/diffusion_quality.py, the accuracy analogue of the KLD workflow: hold
prompt + seed fixed, render a grid with a reference quant (default BF16), then render
each candidate quant and measure drift from the reference. Records mean PSNR + SSIM
(pure-numpy, no skimage/scipy) and optional CLIP text-alignment + image-similarity
(transformers, --clip), plus file size, latency, and peak VRAM, then prints a
quality-vs-cost table and recommends the smallest quant within a quality budget.
--selftest validates the metrics on synthetic images with no GPU or model.
Verified on Z-Image (B200): the table degrades monotonically with quant size
(Q8 -> Q4 -> Q2: PSNR 21.7 -> 15.5, SSIM 0.82 -> 0.61), while CLIP-text stays flat
(~0.34) -- quantization erodes fine detail far more than prompt adherence.
* Studio diffusion (Phase 3): opt-in speed layer (channels_last / compile / TF32)
Add a speed_mode knob (off by default, so the render path stays bit-identical):
default applies channels_last VAE + regional torch.compile of the denoiser's
repeated block where eligible; max also enables TF32 matmul and fused QKV. Regional
compile is gated off for the GGUF transformer (dequantises per-op) and for families
flagged not compile-friendly (a new supports_torch_compile flag, False for Z-Image),
so it activates automatically only once a non-GGUF bf16 transformer is loaded. Speed
optims run before placement/offload, per the diffusers composition order. status now
reports speed_mode + the optims actually engaged.
Verified on Z-Image (B200): default -> ['channels_last'], max -> ['channels_last',
'tf32'], compile correctly skipped for GGUF; generation works in every mode.
121 CPU tests pass.
* Studio diffusion (Phase 2B): opt-in fp8 text-encoder layerwise casting
Add a text_encoder_fp8 knob that casts the companion text encoder(s) to fp8 (e4m3)
storage via diffusers apply_layerwise_casting, upcasting per layer to the bf16
compute dtype while normalisations and embeddings stay full precision. Applied
before placement, gated to CUDA + bf16, best-effort (a failure leaves the encoder
dense). status reports which encoders were cast.
Verified on Z-Image (B200, balanced/group mode where the encoder stays resident):
generation peak VRAM dropped 37% (10840 -> 6791 MB, below the lowest-VRAM offload)
at near-resident speed. It is a memory-vs-quality tradeoff, not free -- ~20 dB PSNR
vs the bf16 encoder, a larger shift than one transformer quant step -- so it is off
by default and documented as such, with the Phase 5 harness to size the cost.
127 CPU tests pass.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Studio diffusion (Phase 2C): NVFP4 text-encoder quant (+ generalise fp8 knob)
Generalise the text-encoder precision knob from a fp8 bool to text_encoder_quant
(fp8 | nvfp4). nvfp4 quantises the companion text encoder to 4-bit via torchao
NVFP4 weight-only (two-level microscaling) on Blackwell's FP4 tensor cores; fp8
stays the broader-hardware path (cc>=8.9). Both are gated, best-effort, and run
before placement; status reports the mode actually engaged. This is the lean
realisation of GGUF-native text-encoder quant: 4-bit on the encoder without the
3045-line port.
Verified on Z-Image (B200, balanced/group where the encoder stays resident), vs the
bf16 encoder: nvfp4 cut generation peak VRAM 48% (10840 -> 5593 MB, the lowest TE
option, below whole-model offload) at near-fp8 quality (16.4 vs 17.1 dB PSNR), and
both quants ran faster than bf16. A memory-vs-quality tradeoff (off by default);
size it per model with the Phase 5 quality harness. diffusion_bench gains
--text-encoder-quant.
129 CPU tests pass.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Studio diffusion (Phase 4): native stable-diffusion.cpp engine for CPU/Mac
Adds the CPU / Apple-Silicon tier of the two-engine strategy, mirroring the
chat backend's llama.cpp shell-out. Diffusers stays the default on CUDA / ROCm
/ XPU; this covers the hardware diffusers serves poorly, consuming the same
split GGUF assets Studio already curates.
- sd_cpp_args.py: pure sd-cli command builder. Maps the family to its
text-encoder flag (Z-Image Qwen3 to --llm, Qwen-Image to --qwen2vl, FLUX.1
CLIP-L + T5), and the diffusers memory policy (none/group/model/sequential)
to sd.cpp's offload flags (--offload-to-cpu / --clip-on-cpu / --vae-on-cpu /
--vae-tiling / --diffusion-fa), so one user knob drives both engines.
- sd_cpp_engine.py: SdCppEngine over a located sd-cli. find_sd_cpp_binary()
with the same precedence as the llama finder (env override, then the Studio
install root, then in-tree, then PATH), an is_available/version probe, and a
one-shot subprocess generate that streams progress and returns the PNG.
runtime_env() prepends the binary's directory to the platform library path
so a prebuilt's bundled libstable-diffusion.so resolves.
select_diffusion_engine() is the pure routing decision (GPU backends to
diffusers, CPU/MPS to native when present).
- install_sd_cpp_prebuilt.py: resolve + download the per-host prebuilt
(macOS-arm64/Metal, Linux x86_64 CPU, Vulkan/ROCm/Windows variants) into the
Studio install root. resolve_release_asset() is a pure, unit-tested
host-to-asset matrix.
- scripts/sd_cpp_smoke.py: end-to-end native generation harness.
Tests (CPU-only, subprocess/filesystem stubbed): 49 new across args, engine,
routing, runtime env, and the installer resolver. Full diffusion suite 166
passing.
Verified on a B200 box: built sd-cli (CUDA) and the prebuilt (CPU) both
generate Z-Image-Turbo Q4_K end to end through SdCppEngine: balanced (group
offload, 5.0s gen), low_vram (full CPU offload + VAE tiling, 13.4s), and the
dynamically-linked CPU prebuilt (50.4s on CPU), all producing coherent images.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Studio diffusion (Phase 4): enforce the sd-cli timeout while reading output
Iterating proc.stdout directly blocks until the stream closes, so a sd-cli that hangs
without producing output (or without closing stdout) would never reach proc.wait and the
wall-clock timeout was silently bypassed. Drain stdout on a daemon thread and wait on the
PROCESS, so the main thread always enforces the timeout and kills a hung process (which
closes the pipe and ends the reader). Add a test that times out even when stdout blocks,
and make the no-binary test hermetic so a host-installed sd-cli can't leak in.
* Studio diffusion (Phase 4) review fixes: sd.cpp installer + engine hardening
- install_sd_cpp_prebuilt: download the release archive with urlopen + an explicit
timeout + copyfileobj (urlretrieve has no timeout and hangs on a stalled socket);
extract through a per-member containment check (Zip-Slip guard); expanduser the
--install-dir so a tilde path is not taken literally; and on Windows CUDA also fetch
the separately-published cudart runtime DLL archive so sd-cli.exe can start.
- sd_cpp_engine: find_sd_cpp_binary honors UNSLOTH_STUDIO_HOME / STUDIO_HOME like the
installer, so a custom-root install is discovered without UNSLOTH_SD_CPP_PATH; start
sd-cli with the parent-death child_popen_kwargs so it is not orphaned on a backend
crash; reap the SIGKILLed child (proc.wait) so a cancel/timeout does not leave a zombie.
- tests: Zip-Slip rejection, normal extraction, studio-home discovery.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Studio diffusion (Phase 4) review round 2: collect sd-cli batch outputs
Codex review: when batch_count > 1, stable-diffusion.cpp's save_results() writes
the numbered files <stem>_<idx><suffix> (base_0.png, base_1.png, ...) instead of
the literal --output path. SdCppEngine.generate checked only the literal path, so
a batch generation would exit 0 and then raise 'no image' (or return a stale
file). generate now returns the literal path when present and otherwise falls
back to the numbered siblings; single-image behavior is unchanged.
Test: a fake sd-cli that writes img_0.png/img_1.png (not img.png) is collected
without error.
---------
Co-authored-by: oobabooga <112222186+oobabooga@users.noreply.github.com>
* MLX CI: find llama-cli where save_pretrained_gguf actually installs it
The GGUF reload step hardcoded the CWD-relative paths llama.cpp/llama-cli and
llama.cpp/build/bin/llama-cli, but save_pretrained_gguf builds and installs llama.cpp
under unsloth_zoo's LLAMA_CPP_DEFAULT_DIR ($UNSLOTH_LLAMA_CPP_PATH, else
~/.unsloth/llama.cpp), so the reload could not find the binary and failed the Mac M1
job with "llama-cli not found". _find_llama_cli now searches that install directory
(and honors the env override) before falling back to the old CWD layout, with a
recursive glob as a last resort. The search is a strict superset of the previous
paths, so it cannot regress a layout that already worked.
* MLX CI: return an absolute llama-cli path from the locator
Resolve the located binary to an absolute path. If UNSLOTH_LLAMA_CPP_PATH is a
relative directory (e.g. "."), Path(".") / "llama-cli" normalizes to the bare name
"llama-cli", and subprocess.run treats a separator-less argument as a PATH lookup
rather than a file to execute, raising FileNotFoundError. resolve() makes the returned
path absolute so it always runs the intended binary.
* MLX CI: give llama-cli EOF on stdin so GGUF reload cannot hang
With the binary now found, the GGUF reload actually invokes llama-cli and it timed
out after 300s generating 24 tokens on a 270m model, which is a stdin block rather
than slow generation: subprocess.run captured stdout/stderr but left stdin inherited,
so -no-cnv still left llama-cli waiting for interactive input. Pass
stdin=subprocess.DEVNULL so it receives an immediate EOF and runs the single prompt to
completion.
---------
Co-authored-by: danielhanchen <michaelhan2050@gmail.com>
Condense the verbose comments and docstrings added by the recent
chat template, GPT-OSS detection, PEFT tensor-parallel, and Studio
inference proxy fixes. Comments and whitespace only; no code changes.
check_js_file put the bundle's KB size inside the finding's check label, which is
part of the baseline match key (package, file, check). When tensorboard's
projector_binary.js grew from 1918 KB to 1933 KB, the reviewed baseline entry stopped
matching and the benign HIGH resurfaced, red-failing the studio and extras
scan-packages shards. Move the size into the evidence field (shown for review, not
matched) and keep the check label constant, then update the one tensorboard baseline
entry to the size-agnostic label. The finding is suppressed again and will not
re-break when the bundle grows by a few KB. Scanner self-tests pass unchanged.
Co-authored-by: danielhanchen <michaelhan2050@gmail.com>
* add models for /update endpoint
* add logic for identifying out of date hf models
* add endpoint for updating hf models
* add relevant field to GgufVariantDetail
* make exception handling better
* add update_available flag for cached_models, and moved /update endpoint from inference -> models
* hook up /update endpoint on the frontend
* implement update scenarios for the model picker
* fix bug where downloaded flag for an older revision was being wrongly set to false
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* fix import and make hf calls async
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* remove has_vision from UpdateRequest
* fix ci
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* clear cancel event before updating gguf variant
* set _cancel_event back if it was set initially
* add hf_token to get_paths_info
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* studio: harden model update endpoint and update checks
- update_hf_model: pass snapshot_download local_dir (local_path is not a
valid kwarg and 500s when updating bicodec audio models)
- get_gguf_variants: wrap the remote update check so a network, rate-limit,
gated, or offline failure degrades to "no update info" instead of failing
the whole variant listing, matching list_cached_models
- add regression tests for both paths
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Studio: HF model update detection and Update action for cached models
Surface an "Update available" cue and a managed Update action for cached
on-device models. /api/hub/update-status compares each cached main GGUF
file's local blobs against the remote main revision using set membership
across all cached revisions, so a repo that was already updated (and still
holds the old snapshot alongside the new one) is not falsely flagged.
The Update action re-downloads through the download manager so it shows in
the Downloads panel with progress and cancel. The frontend wires the Update
button into the GGUF, on-device, and model-selector cards and keeps the
quant label fully visible when the action buttons crowd the row.
Adds regression tests for the multi-revision update check.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Studio: accept force_download kwarg in hf_xet_fallback test double
The download seam now passes force_download to the attempt callable; the _FakeAttempt mock did not accept it, failing 6 tests with TypeError. Add the keyword (default False) so the scripted-results double matches the seam.
* Fix Studio model update regressions
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Address Studio update review feedback
* Address Studio update edge cases
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Share GGUF update status helper
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Fix GGUF update detection and cache cleanup
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Fix cached GGUF update badges
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: shimmyshimmer <107991372+shimmyshimmer@users.noreply.github.com>
Co-authored-by: Etherll <61019402+Etherll@users.noreply.github.com>
Co-authored-by: Lee Jackson <130007945+Imagineer99@users.noreply.github.com>
* fix: keep LoRA reloads working with PEFT 0.19
* test: exercise the PEFT tensor-parallel symbol extractor
* test: prove the full PEFT tensor-parallel seam
* fix: harden PEFT tensor-parallel shims
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* fix: fall back when PEFT tensor-parallel source inspection fails
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Lee Jackson <130007945+Imagineer99@users.noreply.github.com>
Co-authored-by: imagineer99 <samleejackson0@gmail.com>
* i18n: register Japanese language support in messages
* i18n: add Japanese locale support
* Update studio/frontend/src/i18n/locales/ja.ts
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
* Update studio/frontend/src/i18n/locales/ja.ts
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
* test(i18n): add ja locale to parity check
* i18n: fill remaining missing keys for Japanese locale
* i18n: fix terminal string localization in ja locale
---------
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: Lee Jackson <130007945+Imagineer99@users.noreply.github.com>
* fix(studio/llama_cpp): disable trust_env on the loopback health probe
_wait_for_health() polls http://127.0.0.1:<port>/health with the default
httpx trust_env=True, so an ambient HTTP(S)_PROXY in the environment is
applied to the loopback request. A proxy that returns 503 for 127.0.0.1
makes every probe fail, so the loop runs until timeout and Studio load
hangs (trust_env=False returns 200 immediately).
Pass trust_env=False so the local readiness probe never goes through a
proxy. This mirrors the existing trust_env=False handling in the sibling
llama_http / external_provider HTTP clients.
* test(offline_gguf_cache): accept trust_env kwarg in fake_get mock
_wait_for_health now calls httpx.get(..., trust_env=False); update the retry test's fake_get to accept the kwarg so it doesn't raise TypeError.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* fix(studio/llama_cpp): bypass proxies for loopback clients
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* fix(studio/routes): bypass proxies for llama streams
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: wasimysaid <wasimysdev@gmail.com>
* studio: announce Cloudflare tunnel state and warn about public exposure on startup
The startup banner only printed a line when a tunnel URL was up, so a plain
`unsloth studio -H 0.0.0.0` launch silently created a public trycloudflare.com
URL with no indication that Studio had become reachable from the internet. The
only hint at the tunnel was the CLI help, shown when an invalid command was typed.
Make the banner always state the tunnel state for wildcard binds:
- ON: the public URL plus a warning that anyone with it can reach Studio from
outside the network, and that --no-cloudflare keeps it local-only.
- FAILED: requested but did not start (local network only).
- OFF: --no-cloudflare was passed (local network only).
Secure mode keeps its existing wording (the authenticated tunnel is intended and
--no-cloudflare is not valid there). Clarify the --cloudflare help text in both
the argparse and typer definitions. Default behavior is unchanged.
Also surface the state on the `unsloth studio run` banner, which runs the server
with silent=True and prints its own banner: it now calls _print_cloudflare_line
too, so the ON/OFF/FAILED notice and public-exposure warning are no longer
skipped on that path (previously it only echoed the URL when a tunnel was up).
For the OFF and FAILED notices, do not claim "local network only" when the
reachability probe just confirmed the raw port is reachable from the public
internet: --no-cloudflare and a failed tunnel disable only the Cloudflare link,
not the wildcard bind, so the message is reworded to flag the public raw port.
* Fix/adjust Cloudflare banner warnings for PR #6515
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Fix/adjust Cloudflare banner comments for PR #6515
* Fix/adjust IPv6 Cloudflare tunnel gate for PR #6515
* Fix/adjust Cloudflare review comments for PR #6515
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Fix silent run Cloudflare notice
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: wasimysaid <wasimysdev@gmail.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Studio: quick eject from the model selector
Add a one-click eject shortcut to the loaded-model pill so users do not
have to open the picker to unload a model.
- The loaded-status indicator shows a green checkmark at rest and swaps to
a red eject icon on pill hover, with an "Eject model" tooltip. Clicking
it ejects without opening the picker.
- On Device tab now uses the placeholder "Search local models" instead of
"Search Unsloth models".
- The picker's "Eject model" button uses medium font weight.
* Studio: drop unused group/eject marker class on the eject control
* Studio: make the inline eject control valid HTML
The eject shortcut was a focusable span (role/tabIndex) nested inside the
trigger button. A button's content model forbids focusable descendants, so
make it a plain decorative span (aria-hidden, no role/tabIndex) that keeps
the mouse shortcut. Keyboard and screen-reader users eject via the picker's
"Eject model" button.
* Studio: disable the inline eject shortcut on touch devices
On touch (no hover) the red eject icon and title tooltip never reveal, so
tapping the loaded pill could unload the model with no visible affordance.
Add [@media(hover:none)]:pointer-events-none so taps fall through to the
trigger and open the picker; touch users eject from the picker instead.
---------
Co-authored-by: shimmyshimmer <shimmyshimmer@users.noreply.github.com>
Co-authored-by: Wasim Yousef Said <wasimysdev@gmail.com>
* Add whole-document context mode to RAG chat attachments
Thread-attached files are injected in full when they fit a token budget,
instead of only top-K retrieved chunks, so the model reads the entire file
for summarize/reason-over-document requests. Oversized files fall back to
top-K retrieval so the context window is never blown. KB and project
corpora are unchanged (still retrieval).
- core/rag/store.py: all_chunks_for_scope returns every completed-document
chunk for a scope, ordered document-then-index, joined with filename.
- core/rag/tool.py: whole_document_context renders the chunks as the same
<chunk> blocks + citation source-map retrieval produces, returns None
when empty or over budget.
- core/inference/tools.py: build_rag_autoinject tries whole-document first
for thread scopes, falls through to search_for_autoinject otherwise.
- core/rag/config.py: THREAD_WHOLE_DOC + WHOLE_DOC_MAX_TOKENS (env-tunable).
- tests/test_rag_whole_document.py: store ordering, whole-doc render +
budget cutoff, auto-inject whole-doc vs top-K fallback, KB never whole-doc.
* Add scanned-PDF OCR fallback to RAG ingestion
A PDF page with no extractable text layer (a scanned or image-only page)
previously ingested as empty, so image PDFs were invisible to retrieval and
whole-document context. Such pages are now rendered and transcribed by the
loaded vision model during ingestion, so they become searchable and readable
like any other page. This restores OCR for the RAG document flow without a
separate extraction pipeline.
- core/rag/parsers.py: render_pdf_pages renders whole pages (1-based) to PNG.
- core/rag/captioner.py: factor the shared vision call into _vision_complete;
add _ocr_one + ocr_pages (transcribe rendered pages, OCR_MAX_PAGES bound).
- core/rag/ingestion.py: _ocr_scanned_pages runs right after parse, replacing
text on near-empty PDF pages. No-op when OCR is off, no page is scanned, or
no vision model is loaded (degrades like figure captioning).
- core/rag/config.py: OCR_SCANNED, OCR_MIN_CHARS, OCR_MAX_PAGES, OCR_DPI,
OCR_TIMEOUT_S, OCR_MAX_TOKENS (env-tunable).
- tests/test_rag_ocr_fallback.py: page render, ocr_pages gating + cap, scanned
PDF end-to-end OCR into chunks + whole-doc, born-digital skips OCR, disabled
leaves the page empty.
* Broaden OCR prompt to figures/tables and guard against repetition runaway
The OCR prompt now asks the vision model to also transcribe text inside figures,
diagrams, charts and tables, so labels and table cells on scanned pages are
indexed rather than skipped. Verified on real documents that this does not
regress plain-text transcription.
Some vision models loop on sparse images (e.g. a title-only cover) and emit the
same line hundreds of times. _collapse_runaway caps any run of identical
consecutive lines so a pathological page cannot flood the index; legitimate
short repeats (a label appearing a few times) survive. Applied in ocr_pages.
* Restrict whole-document injection to thread attachments only
whole_document_context resolved the combined project+thread scope, so a project
chat (the frontend sends both thread_id and project_id) injected the entire
project corpus in full, contradicting the design that project and KB corpora stay
retrieval-only. A large project corpus could also push the total over budget and
drop a small thread attachment back to top-K.
Resolve the thread scope alone in whole_document_context, and in
build_rag_autoinject only enter whole-doc mode when a thread attachment is present
and no KB is selected (a KB pick is exclusive: search that corpus). Project
sources and KBs keep top-K retrieval. Adds regression tests for the mixed
project+thread payload, the budget isolation, and KB precedence.
* Address review: keep project retrieval, harden budget + OCR guards
Follow-up to the 8-reviewer pass on the whole-document + OCR work.
- Preserve project grounding in project chats. The thread-scope-only fix made
whole-doc exclusive of retrieval, so a thread attachment silently dropped the
project corpus for that turn. build_rag_autoinject now whole-docs the thread
attachment AND retrieves the project sources top-K, merged under one citation
numbering via tool.render_sources. KB selection stays exclusive.
- Budget: a NULL/zero token_count no longer bypasses the cap (length-based
fallback in _row_token_count), so a malformed huge doc can't inject in full.
- OCR runaway guard: _collapse_runaway now also caps each distinct line at a
generous total across the page (not just consecutive), bounding the
interleaved/alternating loops weak models emit; blank-line floods collapse too.
- OCR: warn when a scanned PDF exceeds OCR_MAX_PAGES (pages past the cap stay
untranscribed) instead of silently dropping them.
- Document the known limits: OCR'd pages have no PDF highlight regions; vision
models need a micro-batch >= image tokens (Gemma-family) or the server aborts.
- Tests for project-retrieval composition, NULL-token budget, and interleaved
runaway; drop the now-superseded exclude-project test.
* Add OCR toggle to RAG retrieval settings
Make scanned-PDF OCR user-controllable per upload instead of only via the
RAG_OCR_SCANNED config default. The retrieval settings panel gains an OCR
scanned pages switch (persisted in localStorage, on by default); the chosen
value is read fresh at upload time and sent with each document upload.
Backend: the three upload routes accept an optional ocr form field and pass it
through start_ingestion to _ocr_scanned_pages, which now treats None as use the
config default and an explicit bool as an override. The on/off policy lives only
in _ocr_scanned_pages now, so ocr_pages no longer re-checks the config (that
double gate would have blocked a per-upload ocr=True while the default was off).
Tests cover both override directions (force on while config off, force off while
config on).
* Add "Describe figures & charts" toggle with chart-aware captions
Surface RAG figure captioning as a user control and make it actually useful for
graphs and plots. The figure detection already clustered vector drawings and
raster images into regions and rendered them, but captioning was off by default,
had no UI, and used a thin generic prompt.
Accuracy: the caption prompt now asks for chart type, axis titles and units,
legend or series, salient trends and readable values, and table columns, while
forbidding invented numbers. The token budget is configurable (CAPTION_MAX_TOKENS)
and captions pass through the same runaway guard as OCR so a looping vision model
cannot flood the index.
Control: a per-upload caption override threads from the three upload routes through
start_ingestion and _run, with the on/off policy single-sourced in _run (caption
self-gating removed from caption_images, mirroring the OCR change) so a force-on
override works when the config default is off. The frontend adds a "Describe
figures & charts" switch in the retrieval settings, persisted in localStorage and
sent with each upload. Default on; it is a no-op without a vision model and bounded
to CAPTION_MAX_IMAGES figures per document.
Tests cover the new caption_images contract, the runaway guard on captions, the
chart-aware prompt and token budget (and that OCR keeps its own prompt and budget),
and both override directions end to end through ingestion.
* Generalize figure understanding: transcribe-first prompt + high-DPI tiling
Make figure/chart description work across any visual and any model strength, not
just a strong VLM on simple figures. Two changes, validated by a recall benchmark
on authoritative documents (ResNet/Attention papers, USDA, UN UDHR).
1. Transcribe-first caption prompt. The caption now asks the model to transcribe
every visible label verbatim (titles, axis labels and units, legends, every
box/node/arrow label, table cells, equations) and then add a one-line summary,
instead of only describing the figure. Transcription is the most model-robust
visual task, so weak models that cannot reason about a chart still recover its
labels.
2. High-DPI tiling of figure pages. Figure-bearing pages are rendered as an
overlapping grid of high-DPI tiles (plus a full-page pass for context); each
tile is transcribed, then merged and de-duplicated. This keeps small diagram
labels legible and covers every sub-figure without relying on exact region
detection, which previously missed sub-figures and small labels.
Supporting changes: figure render DPI 130 -> 200 with a clip margin so edge labels
are not lost; vision calls are deterministic (temperature 0) so transcription does
not randomly drop labels; the repetition guard now applies to captions too. New
config knobs: FIGURE_DPI, FIGURE_MARGIN_FRAC, FIGURE_TILE_ROWS/COLS, FIGURE_TILE_
OVERLAP, FIGURE_FULLPAGE, CAPTION_MAX_PAGES, larger CAPTION_MAX_TOKENS, and
CAPTION_MAX_IMAGES as a per-document tile budget.
Measured figure context recall (per-label, dense academic figures):
Qwen2.5-VL: 0.50 -> 0.83 (overall 0.81 -> 0.94)
Gemma-4-E2B (weak): ~0 with loops -> 0.83 (overall 0.91)
Born-digital text and scanned-page recall are unchanged (no regression).
parsers gains _figure_boxes (shared detection), pages_with_figures, and
render_pdf_figure_tiles; captioner gains merge_page_captions and a temperature
parameter; ingestion routes figure captioning through the tiled path.
* Fix RAG review issues: whole-doc budget pre-check, figure gating, empty re-ingest, vision auth
Whole-document context now runs a cheap token-sum pre-check (store.scope_token_estimate)
before hydrating every chunk's text, so an attachment that cannot fit the budget is
rejected without loading the whole corpus into memory. The estimate mirrors
all_chunks_for_scope's filter and the per-row token-count fallback exactly.
Ingestion skips all figure work (PDF rasterization and detection, not just the caption
call) unless a vision model is loaded, so a text-only deployment pays nothing. When OCR
is enabled, scanned/image-only pages are excluded from figure tiling since OCR already
transcribes them whole, avoiding double vision work and overlapping index entries; a
scanned figure page is still tiled when OCR is off.
start_ingestion no longer dedupes forever to a prior ingest that produced zero chunks
(e.g. a scanned PDF uploaded before a vision model was loaded): the empty record is
dropped and the content is re-ingested.
Vision OCR and caption requests now send the backend Authorization header, so they
match the chat endpoint and do not 401 under direct-stream (--api-key) mode.
Adds tests for the budget estimate, scanned-page exclusion, the vision-model gate, the
empty re-ingest path, and the auth-header passthrough.
* Trim RAG vision-ingestion comments and docstrings
Tighten the verbose multi-line docstrings and comments added across the RAG vision
ingestion work (captioner, config, parsers, ingestion, store, tool, build_rag_autoinject,
the RAG tests, and the chat-store/upload-hook frontend toggles) to one or two lines while
keeping their intent. No code changed: verified comment/docstring-only against the prior
commit, and the RAG test suite still passes.
* Fix figure-tiling exclusion and client dedupe for re-ingestable docs
Figure tiling now excludes only the pages OCR actually transcribed, not every
text-less page. _ocr_scanned_pages returns the set of pages it OCR'd, and _run passes
that to pages_with_figures as exclude_pages (replacing the ocr_on-keyed min_text_chars
heuristic). A scanned page that OCR skipped (past OCR_MAX_PAGES, or whose OCR returned
empty) is no longer dropped from captioning, so a chart on such a page still gets a
caption.
The document panel's upload dedupe no longer skips re-selecting a file whose only
matching doc completed with zero chunks. Such a doc is re-ingestable (e.g. a scan
attached before a vision model loaded), and the backend re-ingests on the same content
hash, so the client must let it reach the backend; healthy or still-indexing docs are
still skipped. The SSE complete frame's chunk count is recorded on the doc so the
check is exact.
Adds a regression test for the un-OCR'd scanned figure page and updates the
pages_with_figures test to the exclude_pages interface.
* Address review findings: whole-doc budget guard, job numChunks, dead code, upload cap
whole_document_context now treats a non-positive max_tokens as "never inject" instead
of injecting the whole corpus unbounded, so RAG_WHOLE_DOC_MAX_TOKENS=0 tightens rather
than disables the budget (the real off switch stays RAG_THREAD_WHOLE_DOC=0).
The job-status endpoint and get_job_status now expose num_chunks (joined from the
document), and the upload hook threads it through the SSE-fallback completion paths
(reconcile + poll). Previously a document that finished via the connection-cap fallback
had no chunk count client-side, so the re-ingest dedupe wrongly treated it as empty and
re-uploaded it. IndexJob/JobEvent gain the field and the untyped cast is dropped.
Removes the dead render_pdf_figures function (superseded by the tiling path), its test,
and the unused FIGURE_MARGIN_FRAC config knob.
Adds an upload size cap (RAG_MAX_UPLOAD_BYTES, default 200 MB; 413 on exceed with the
partial file cleaned up) so a pathological file can't drive unbounded parse + vision
work. render_pdf_figure_tiles clamps rows/cols to >= 1 (no ZeroDivisionError on a
misconfigured grid). Captioning progress is reported after OCR so the bar is monotonic.
sqlite connections set busy_timeout=5000 so a long figure/scan ingest holding its
connection doesn't make a concurrent ingest/read fail with "database is locked".
Adds tests for the non-positive budget, the zero-grid clamp, job-status num_chunks, and
the oversize-upload rejection.
* Extract PDF text as layout-aware Markdown via pymupdf4llm
parsers._pdf now extracts each PDF page as Markdown with pymupdf4llm.to_markdown
(page_chunks=True) instead of flat page.get_text("text"), so tables, headings and lists
keep their structure in the indexed chunks and retrieve far better (a table's cells stay
associated with their row instead of flattening into a token stream). Gated by
RAG_PDF_MARKDOWN (default on); falls back to plain PyMuPDF text when the toggle is off,
pymupdf4llm is missing, extraction fails, or a page yields no Markdown. The scanned-page
OCR and figure-tiling passes operate on rendered pixels and are unaffected; docx/html/txt
keep their existing extractors.
The preview-highlight locator already strips Markdown punctuation when building anchors;
it now also splits anchor tokens on pipes so a Markdown table row still anchors to the
raw PDF word stream.
Declares pymupdf4llm as a studio/RAG dependency (was only transitively present via the
data-designer plugin). Adds parser tests (Markdown table reaches the page text, the
plain-text fallback, the missing-lib fallback) and a locator test for table-pipe anchoring.
* Pin pymupdf4llm to 0.3.4 so the package scan does not pull onnxruntime
The lockstep pymupdf4llm 1.27.x line makes pymupdf-layout a hard dependency,
which in turn pulls onnxruntime (plus numpy/networkx/protobuf). The security-audit
pip scan-packages job resolves requirements --with-deps, so adding pymupdf4llm to
no-torch-runtime.txt and studio.txt surfaced onnxruntime's un-baselined CRITICAL
finding and flipped the hf-stack shard from pass to fail.
pymupdf4llm 0.3.x keeps pymupdf-layout behind an optional [layout] extra, so a plain
install resolves to pymupdf + tabulate only and never touches onnxruntime. 0.3.4
requires pymupdf>=1.27.1, satisfied by our pinned pymupdf==1.27.2.3, and to_markdown
(page_chunks=True) produces equivalent layout-aware Markdown on real PDFs (verified on
the Attention, ResNet and USDA documents). Production already installs these files
--no-deps, so onnxruntime was never shipped at runtime; this only fixes the scanner.
The parser test now asserts Markdown markup (heading or table pipes) rather than table
pipes specifically, since 0.3.4 emits a heading but not a pipe table on the tiny
borderless synthetic fixture; both markers are absent from the plain-text fallback.
* Fix RAG whole-doc review findings
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Address RAG whole-doc review follow-ups
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Address RAG review follow-up edge cases
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Reserve image budget for whole-document RAG
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: danielhanchen <michaelhan2050@gmail.com>
Co-authored-by: Lee Jackson <130007945+Imagineer99@users.noreply.github.com>
Co-authored-by: wasimysaid <wasimysdev@gmail.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Studio: wire imatrix GGUF option and FP8/NVFP4 compressed export into the export UI
GGUF export gains an importance-matrix toggle. When enabled it auto-downloads the
upstream Unsloth imatrix for the base model (or uses a custom path), which unlocks
the IQ low-bit quants iq2_xxs, iq2_m, iq3_xxs and iq4_xs. Merged export gains an
FP8 / NVFP4 compressed-tensors precision selector that runs llm-compressor for vLLM.
Backend threads imatrix_file through routes -> orchestrator -> worker -> export_gguf
(both the local save and the hub push), and maps the new compressed format_type
values onto the fp8/nvfp4 save_method, reporting the "<dir>-<suffix>" sibling output
directory. Frontend adds the imatrix Switch on the GGUF card and a merged precision
picker on the merged card, threaded through the export runtime store.
Depends on unslothai/unsloth#6706 (save.py imatrix_file and compressed-tensors
export) and unslothai/unsloth-zoo#839 (quantize_gguf imatrix flag).
* Studio export: guard imatrix/compressed against older unsloth builds and force imatrix for IQ quants
Addresses review feedback on the export wiring:
- GGUF: pass imatrix_file only when set, so a plain no-imatrix export (e.g. Q4_K_M) no
longer fails with an unexpected-keyword error against an unsloth build that predates the
imatrix_file parameter. When imatrix is requested but unsupported, return a clear
upgrade message instead of a TypeError.
- Merged: gate FP8/NVFP4 compressed-tensors export on the installed unsloth actually
supporting it, returning a clear message rather than a cryptic save_method failure.
- Frontend: IQ quants (iq2_xxs, iq2_m, iq3_xxs, iq4_xs) are imatrix-only, so force the
imatrix on when one is selected and lock the toggle, instead of submitting an IQ quant
with no imatrix that llama.cpp would reject.
Extends the backend tests for the new capability guards and the conditional kwarg wiring.
* Studio: upload compressed merged models to the Hub without recompressing
For an FP8/NVFP4 Hub export the model is already produced locally in the "<dir>-<suffix>"
output. Uploading it directly with HfApi.upload_folder (mirroring export_base_model) avoids
re-running the expensive compressed-tensors quantization a second time inside
push_to_hub_merged, which for NVFP4 also re-runs calibration and risks OOM. Falls back to
push_to_hub_merged when there is no local compressed output to reuse.
* Add FP8/FP4 compressed export to save_pretrained_merged
Adds compressed-tensors export (for vLLM) to save_pretrained_merged /
push_to_hub_merged via llm-compressor, alongside the existing lora /
merged_16bit / merged_4bit / gguf / torchao paths:
model.save_pretrained_merged("model", tokenizer, save_method="fp8")
Supported save_method values: fp8 (FP8_DYNAMIC), mxfp4, nvfp4 (W4A4) and
mxfp8. The LoRA is merged to 16bit at save_directory, then a quantized
checkpoint is written to save_directory + "-<fmt>". nvfp4 needs a small
calibration set (defaults to ultrachat, overridable via calibration_dataset).
Notes:
- llm-compressor is installed lazily on first use, pinning the current torch
and transformers via a constraints file so they are not upgraded (a plain
install pulls transformers>=5 and breaks Unsloth).
- Quantization runs in a separate process (unsloth/_compressed_quantize.py,
launched by file path) so Unsloth's transformers attention patches do not
interfere with the forward llm-compressor runs during calibration, mirroring
how GGUF export shells out to llama.cpp.
- mxfp8 needs a newer llm-compressor (transformers>=5); it is recognised and
raises a clear error until that stack is available.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Address review: main-process guard, calibration subsampling, tokenizer + dtype handling
- Route the 16bit merge through unsloth_generic_save for both LoRA and full
finetuned models, so non-PEFT models are written in 16bit consistently
instead of saving the original (possibly quantized) weights directly.
- Honor is_main_process: only the main process quantizes and writes the
compressed output, so distributed ranks do not race on the same dirs.
- Subsample an in-memory calibration Dataset before save_to_disk so large
training sets are not fully copied to a temp dir.
- Tolerate a missing tokenizer in the converter (data-free exports); still
require one for calibration based schemes.
- Open config.json via a context manager in both files.
- Drop the redundant nvfp4 entry from the unsupported-name check (fp4 covers it).
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Add direct LoRA to GGUF export and harden FP8/FP4 compressed export
- Run llm-compressor install and scheme check before the 16bit merge so
unsupported schemes (e.g. mxfp8) fail fast without writing a checkpoint
- Only the main process installs, merges, quantizes and uploads; isolate
hub pushes to a temp dir and clean all temp dirs in a finally
- Forward standard save kwargs (state_dict, max_shard_size, ...) to the merge
- Fall back to the first dataset split for Hub calibration ids
- Export LoRA adapters to GGUF via convert_lora_to_gguf.py: modernize
save_pretrained_ggml/push_to_hub_ggml and add save_method="lora" to
save_pretrained_gguf/push_to_hub_gguf; resolve base from the adapter config
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Fix LoRA GGUF shell-injection test and compressed export trailing-slash path
- Update tests/saving/test_save_shell_injection.py for the new delegation: the
LoRA to GGUF conversion now lives in _unsloth_save_lora_gguf, so assert it
passes argv as a list with no shell=True and that the legacy ggml wrappers
delegate to it instead of calling subprocess.Popen directly
- Normalize the local save_directory before building the "<dir>-<fmt>" sibling
so a trailing slash no longer nests the compressed output inside the 16bit dir
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Polish FP8/FP4 and LoRA GGUF export after review
- Warn (not silently downgrade) when an explicit quantization_method is not a
valid LoRA GGUF outtype; default stays f16
- Correct the inference hardware note: MXFP8 is 8-bit (cc >= 8.9), only FP4
needs Blackwell for full activation quantization
- Document that a local fp8/fp4 save keeps the 16bit merge at save_directory
and writes the quantized checkpoint to save_directory + "-<fmt>"
* Use sequential calibration pipeline and validate Hub access early
- nvfp4 calibration no longer forces the memory-hungry "basic" pipeline. The
quantization runs in a clean subprocess, so llm-compressor's default
sequential pipeline (layer-by-layer onloading) works and lets large models
that do not fit at once still calibrate; fall back to "basic" only if tracing
fails
- For push_to_hub compressed exports, create/validate the repo up front so a bad
token or denied repo fails before the merge and quantization instead of after
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Harden compressed export: explicit sequential pipeline, base-tokenizer calibration, GPU memory
- nvfp4 calibration now passes pipeline="sequential" explicitly (layer-by-layer
onloading) instead of relying on the inferred default, with a "basic" fallback
- Calibration datasets with a messages column no longer require a chat template:
base / non-chat tokenizers fall back to concatenating message contents
- Free the in-memory model's CUDA memory before the quantize subprocess loads its
own copy from disk (best-effort, single-device non-quantized only; restored
afterward), so a single GPU need not hold two copies at once
- Create the calibration temp dir in the system temp location instead of next to
the save directory, avoiding stray dirs in the workspace
* Free the failed calibration model before the basic-pipeline retry
In the sequential -> basic NVFP4 fallback, release the partially-processed model
and clear the CUDA cache before loading a fresh copy, so the retry does not
transiently hold two model copies on the GPU.
* Harden calibration data handling and compressed-export edge cases
- Calibration messages without a chat template now handle multimodal (list)
content, None content, and null message rows instead of crashing on join
- Raise a clear error when the calibration dataset is empty after subsampling
- Reset llm-compressor's global session before freeing the model in the
sequential -> basic NVFP4 fallback, so the old model is actually released
- LoRA GGUF export accepts a single-element list quantization_method
- Attach datasets metadata to the pushed repo on compressed hub exports
- Warn (instead of silently) if the model cannot be restored to its device
- Raise a clear error if the LoRA base model id cannot be determined
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Handle DatasetDict calibration, MoE routers, and MTP models in compressed export
- Reduce an in-memory DatasetDict calibration set to a single split before row
subsampling, so save_to_disk does not copy every split to the temp dir
- For MoE models, keep the router/gate unquantized and pass
moe_calibrate_all_experts so every expert is calibrated
- Warn when a model carries MTP / speculative-decoding tensors that the
compressed export does not include
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Support many more compressed-tensors schemes and address review
- Expand save_method to cover the full set of compressed-tensors preset schemes:
FP8 (dynamic/static/block), INT8, W8A8, W8A16, W4A16(+asym), W4A8, W4AFP8,
MXFP4(+A16), NVFP4(+A16), plus the gated MXFP8; calibration is used only for
the static-activation schemes (FP8 static, NVFP4)
- Broaden the near-miss save_method error to cover int/w-prefixed names
- MoE: also keep the Qwen shared-expert gate unquantized
- Strip non-model-input columns from already-tokenized calibration data so the
collator does not choke on a leftover messages column
- Forward the Hub token to the LoRA converter and the quantize subprocess so
gated/private base models and calibration datasets work without a global login
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Collapse compressed-tensors export help line so ruff-format converges
The print line in print_quantization_methods needed two ruff-format passes to
reach a fixpoint (merge implicit string concat, then collapse the single-arg
print). pre-commit.ci applies one pass per run, so it kept reformatting. Land
the converged single-line form directly.
* Add CPU-only regression tests for the export API
Cover all export paths without a GPU, for slow CPU-only CI:
- pure-function checks of the compressed-tensors scheme registry and save_method
normalization (aliases, calibration flags, near-miss errors)
- AST checks that every merged saver dispatches compressed export, the GGUF savers
expose the lora branch, torchao routes PTQ/QAT, the public methods stay attached,
and the export subprocesses remain shell-safe (argv list, sys.executable, no shell)
- monkeypatched dispatch checks that fp8/nvfp4/merged_16bit, the LoRA-GGUF outtype
resolution, and torchao PTQ/QAT reach the right helper with the right arguments
* Run the CPU-only export tests in consolidated CI
tests/saving is --ignored by the Repo tests (CPU) job, so the new GPU-free export
tests are added by path to consolidated-tests-ci.yml (collection sanity + Bucket-A run),
alongside the existing CPU saving tests, so they actually execute on CPU CI.
* Add GPU GGUF export + llama-cli inference smoke test
tests/saving/test_gguf_export_and_inference.py: skipif no CUDA. Trains a tiny
phrase-imprinting LoRA, exports a full-model q8_0 GGUF (merge -> convert_hf_to_gguf
-> llama-quantize), asserts a valid GGUF (magic + size), and - when a llama-cli
binary is available - runs one bounded generation (byte cap + watchdog kill) and
asserts the trained phrase round-trips through HF -> GGUF -> quantize -> inference.
The llama-cli step skips gracefully since the export only builds llama-quantize.
* Fix variant mismatch in compressed (FP8/FP4) export
save_pretrained_merged(..., save_method=fp8/nvfp4, variant=...) forwarded
the variant into the intermediate 16bit merge, so Transformers wrote
variant-named shards (model.<variant>.safetensors). The converter
subprocess then reloaded that directory with the default weight filenames,
so the compressed export failed after doing the merge.
Pop the variant out of the intermediate merge (internal staging that the
subprocess reloads with default names) and forward it via --variant so it
is applied to the final compressed checkpoint instead. Add a CPU AST guard
for the contract.
* Harden export paths from review
- install_llm_compressor: fall back to uv pip when this interpreter has no
pip seeded (uv-created/relocatable venvs), instead of failing with
No module named pip.
- LoRA GGUF export: if convert_lora_to_gguf.py is missing (a prebuilt or
reused CWD llama.cpp install carries binaries but not the converter
script), force a dedicated source checkout that ships it.
- push_to_hub_gguf(save_method=lora): return on non-main ranks, matching the
local save_pretrained_gguf lora branch, so only rank 0 converts/uploads.
- compressed export VLM detection: require a vision_config or a
ForVisionText2Text architecture; a bare *ForConditionalGeneration also
matches text seq2seq models (T5/BART/Whisper) and is no longer treated as
a VLM on its own.
- GGUF GPU smoke test: drop SFTConfig(max_length=1024), which raises under
newer TRL padding-free training; length enforcement is not needed here.
* Add imatrix option to GGUF export, enabling IQ low-bit quants
save_pretrained_gguf / push_to_hub_gguf gain imatrix_file:
None -> no imatrix (unchanged)
'/path' -> pass to llama-quantize --imatrix (a *.gguf_file is renamed to *.gguf)
True -> download the upstream unsloth/<base>-GGUF imatrix (imatrix_unsloth.dat or
.gguf_file), raising a clear error if none exists
An importance matrix unlocks the IQ low-bit quants (iq2_xxs, iq4_xs, ...), which were hard
disabled before. They are gated: requesting one without an imatrix raises a clear error.
- _resolve_imatrix_file resolves path/True (PEFT base first, normalized via get_model_name,
derives unsloth/<base>-GGUF, copies out of the HF cache before renaming *.gguf_file).
- IMATRIX_QUANTS registry replaces the old commented-out IQ entries; save_to_gguf accepts a
resolved imatrix and threads it into the quantize calls.
- The --imatrix flag is emitted by unsloth_zoo's quantize_gguf (companion change). save.py
fails fast with an upgrade hint if the installed unsloth_zoo lacks the imatrix kwarg.
Tests: tests/saving/test_imatrix_export.py (CPU: resolution, repo derivation, IQ gate,
--imatrix wiring) wired into CI; tests/saving/test_gguf_export_and_inference.py extended with
GPU iq2_xxs/iq4_xs export + inference. Verified end to end on Llama-3.2-1B: imatrix
auto-downloaded, iq2_xxs/iq4_xs exported and run via llama.cpp.
Note: requires the companion unsloth_zoo quantize_gguf imatrix change.
* Address imatrix/compressed review feedback: unsloth org GGUF repo, fail-fast, calibration split
- imatrix auto-resolve (imatrix_file=True): derive the upstream repo as unsloth/<base>-GGUF
instead of <org>/<base>-GGUF, so official bases (e.g. meta-llama/Llama-3.1-8B-Instruct) find
the matching Unsloth GGUF imatrix repo rather than failing on a nonexistent meta-llama/...-GGUF.
- Resolve/validate the imatrix before the 16-bit merge in save_pretrained_gguf, so a bad path or
an unavailable upstream imatrix fails fast instead of after a long, multi-GB merge.
- Compressed calibration: when a Hub dataset has no "train" split, resolve the first split name
and slice it, instead of materializing the whole dataset just to take num_samples rows. Keeps
the original materialize-then-subselect path as a last resort.
Tests: add unsloth/<base>-GGUF mapping for an official base id, and create the imatrix file in the
quantize_gguf flag test (quantize_gguf now validates the imatrix exists).
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Studio: name the missing extractor when a Recipes upload fails
A missing optional dependency (pymupdf4llm for PDF, mammoth for DOCX) was
reported as a generic "Text extraction failed", which gives the user nothing to
act on. Catch ImportError and surface the package name instead.
* Studio: narrow missing extractor error handling
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: wasimysaid <wasimysdev@gmail.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* (feat) Add project names to studio training runs to avoid models being overwritten when doing similar training runs
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Update studio/frontend/src/features/export/export-page.tsx
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
* Update studio/frontend/src/features/export/export-page.tsx
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
* Update studio/frontend/src/features/export/export-page.tsx
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* better project name sanitization, removed duplicated project name normalization
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* implement checkpoint scanning utilities and tests for base model inference
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Guard project_name against null and use leading important modifiers
* Fix/adjust training project names for PR #6512
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Fix/adjust training project names for PR #6512
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Address project-name review feedback
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Show project names in training recents
* Keep GGUF export directories source-specific
---------
Co-authored-by: NZ-Linix <nz-linix@outlook.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: NZ-Linix <linus.ordowski@outlook.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: wasimysaid <wasimysdev@gmail.com>
* Speed up Studio desktop startup
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Address Studio startup review findings
* Keep orphaned run cleanup before readiness
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Improve code quality & performance: fix typos, compile regex & cache fields
- Fix typos across core files (repeatted → repeated, splitted → split, etc.)
- Compile regex patterns once as class attributes in TextPreprocessor
- Cache text fields/columns in RawTextDataLoader
- Improve comments (re-use → reuse)
* Use immutable raw text field constants
---------
Co-authored-by: imagineer99 <samleejackson0@gmail.com>
Co-authored-by: Lee Jackson <130007945+Imagineer99@users.noreply.github.com>
Local models in the Studio Hub tab (Custom folders, LM Studio, and
Local models sections) did not reveal their on-disk path on hover,
unlike the Fine-tuned rows which already do. Each of these rows maps
over a LocalModelInfo with a required path, so pass tooltipText built
from the model name and path via a small shared localPathTooltip
helper, matching the existing FT-row tooltip format.
Refs #6382
Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com>
Co-authored-by: Lee Jackson <130007945+Imagineer99@users.noreply.github.com>
* cascade user message deletion to include assistant reply
* Fix comment typo in delete-thread-message
---------
Co-authored-by: Daniel Han <danielhanchen@gmail.com>
FalconH1RMSNormGated is imported from transformers but never referenced in
unsloth/models/falcon_h1.py. The unused hoist trips the import-hoist lint gate
on the merge commit of every open PR (the gate lints PR-head merged into main),
so clearing it here unblocks those PRs.
* Studio: stop handing CI/user secrets to downloaded llama.cpp binaries
The macOS prebuilt path installs llama.cpp from the unslothai/llama.cpp
fork's latest (unpinned, mutable) release and then executes the
downloaded llama-server / llama-quantize binaries during install-time
validation. binary_env() built that child environment from a full
os.environ.copy(), so a compromised or tampered prebuilt would inherit
every secret in the process: HF_TOKEN and the workflow GitHub tokens in
CI, and HF / cloud credentials for end users running install.sh /
setup.sh.
We publish prebuilts daily, so pinning a release tag is not workable.
Instead, neutralise the impact: these binaries have no reason to read any
token, so strip secret-bearing variables (exact names plus
TOKEN/SECRET/PASSWORD/CREDENTIAL/PRIVATE_KEY/API_KEY markers) before
handing the env to a downloaded binary. The installer's own GitHub and
Hugging Face API calls read os.environ directly, so authentication and
release-API rate limiting are unaffected; PATH, LD_LIBRARY_PATH,
DYLD_LIBRARY_PATH and CUDA/ROCm vars are preserved. One change covers the
install-time validation path for all six macOS workflows and end users.
Follow-up (separate, sequenced): publish build-provenance attestations
from the fork's prebuilt workflows and verify them in CI, so a forged
release is rejected rather than merely starved of secrets.
* Strip KUBECONFIG, SSH_AUTH_SOCK, and PASSPHRASE-marked vars from binary env
Extend the deny-list per PR review: KUBECONFIG and SSH_AUTH_SOCK are
credential pointers/capabilities a downloaded binary never needs, and a
PASSPHRASE marker catches SSH_PASSPHRASE / GPG_PASSPHRASE. Tests updated.
* Studio: also scrub proxy/index env vars and URL-embedded credentials before running prebuilt binaries
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Scope mlx-ci secrets to the install + download commands for PR #6696
Drop the ambient step-level env block and pass GH/GITHUB/HF tokens only
on the installer and GGUF-download commands, so the directly invoked
llama-quantize / llama-server smoke runs see no secrets. The installer
still reads tokens from os.environ for the releases API and probe fetch.
* Trim verbose comments around the secret-env scrubber for PR #6696
Comment-only: condense the block comments added across this PR. Logic
unchanged (comment_tools.py check confirms code-only signature equal).
* Redirect HOME / cache pointers to an empty dir for prebuilt binaries (PR #6696)
Address Codex P2: stripping token env vars still let a tampered binary
read on-disk token stores (~/.cache/huggingface/token, ~/.aws/credentials,
~/.config/gh) through $HOME and the cache/config pointers. Point HOME plus
the HF / XDG / Windows home pointers at a single empty throwaway dir for
the downloaded-binary env. Defense in depth: a binary resolving the real
home via getpwuid is out of scope and needs OS sandboxing.
* Close residual credential-probe gaps for PR #6696
Address the latest Codex review:
- Strip token-only URL userinfo too (scheme://ghp_token@host), not just
the user:pass form.
- Redirect HOMEDRIVE/HOMEPATH alongside USERPROFILE so a Windows binary
cannot reconstruct the real profile from %HOMEDRIVE%%HOMEPATH%.
- Drop explicit credential-file pointers (NETRC, PIP_CONFIG_FILE,
DOCKER_CONFIG, GIT_CONFIG_GLOBAL) that live outside HOME.
- Probe ldd with a secret-free env: linux_runtime_dirs ran ldd on the
untrusted prebuilt with the inherited os.environ, and ldd may execute
the binary, so it could observe HF_TOKEN/GITHUB_TOKEN during the probe.
Factored the shared scrub into secret_free_environ().
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Separate token-bearing install from binary smoke; drop CI command files (PR #6696)
Address the two P1s in the latest review:
- mlx-ci: GitHub bakes secrets into the run-script text, so inline token
assignments in a step that later runs the prebuilt let a tampered binary
read them from the script. Split into a token-bearing install + download
step that never launches a binary, and a secret-free smoke step that runs
llama-quantize / llama-server.
- secret_free_environ now drops the GitHub Actions command files
(GITHUB_ENV, GITHUB_PATH, GITHUB_OUTPUT, GITHUB_STEP_SUMMARY, BASH_ENV) and
the smoke step unsets them, so a tampered prebuilt cannot inject PATH/env
into the later token-bearing MLX steps.
* Run the prebuilt smoke last, after all token-bearing steps (PR #6696)
Address the P1 workspace-poisoning vector: even with no secrets in its env,
a tampered prebuilt could edit the checkout or installed modules, and the
later HF_TOKEN MLX steps would then execute that poisoned code on push
builds. Move the prebuilt install + smoke to the end of the job so the
untrusted binary runs after every token-bearing step, leaving nothing for it
to corrupt. The MLX GGUF reload uses a source-built llama-cli, not this
prebuilt, so nothing depends on the earlier position.
* Trim comments around the secret-env scrubber and prebuilt CI steps (PR #6696)
Comment-only: condense the security-rationale block comments and merge the
duplicated prebuilt-step description in mlx-ci. Logic unchanged
(comment_tools.py check confirms the code-only signature is equal; install
suite still passes).
* Authenticate the GGUF export release-API lookup with the read-only GITHUB_TOKEN (PR #6696)
* Rename env scrubber off the secret-named identifier CodeQL flags as a clear-text sink (PR #6696)
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Studio: restore tensor parallelism for vision/mmproj GGUFs
#6416 disabled --split-mode tensor for any GGUF that ships an mmproj projector to
dodge a GGML_ASSERT crash (#6415) seen on an older llama.cpp build with consumer
Blackwell (sm_120). The blanket skip silently dropped tensor_parallel=true for
every multimodal/MTP GGUF (e.g. Qwen3.6-35B-A3B-MTP); on hardware where the model
fits on one GPU the load then collapsed to a single GPU. mmproj + --split-mode
tensor works on current builds (verified end to end on B200/sm_100), so the skip
was disabling a working configuration.
Make the vision skip self-healing per binary:
- attempt tensor for vision models by default
- skip upfront only on a binary already seen to abort on tensor + mmproj this
session (_vision_tensor_split_aborts), recorded when such a launch crashes at
startup (_record_vision_tensor_split_abort). Process scoped, so a studio update
re-probes the new build. The route-level layer-split fallback stays the net.
- add _select_gpus(min_gpus=...) so a downgraded tensor request can keep multiple
GPUs instead of collapsing to one (default 1, no behavior change).
Add tests/test_tp_vision_regression.py: an AST allowlist guard over the
tensor_parallel drop sites (which would have flagged #6416), plus cache and
_select_gpus coverage. No GPU required.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Studio: address review on vision tensor-parallel self-healing
Three fixes from the PR review:
- Record a vision-tensor abort only after every startup retry fails. The first
version cached the binary on the first spawn crash, which on every build
(including capable ones) is the benign --fit step abort that the existing
--fit off retry resolves. That poisoned the cache so the next vision load in
the same process skipped tensor. Recording now happens at the post-retry
failure block (after fit-off, flash-attn-off and MTP-drop), so a binary that
actually works is never cached.
- Gate the record on the tensor/mmproj crash signature: a hard signal fault
(_is_signal_crash) with no non-tensor cause (_output_has_nonprojector_diagnostic
excludes OOM and unknown-arch), so an OOM, bad extra args, or MTP/flash-attn
crash no longer marks an otherwise capable binary incompatible.
- Preserve the multi-GPU request on the cached downgrade. The vision gate now
raises _layer_min_gpus to the visible GPU count and threads it through the
layer-split GPU selection (_select_gpus min_gpus and the subset loops), so a
downgraded tensor request still spreads across GPUs instead of collapsing to a
single card the model happens to fit.
Verified two vision+tensor loads in one backend process both tensor-split across
4 GPUs (the benign fit abort no longer poisons the cache). Tests updated.
* Studio: harden vision tensor-parallel self-healing (review round 2)
Address the second review round on the vision/mmproj tensor-parallel fix:
- Preserve vision on the first load: a --split-mode tensor + --mmproj
GGML_ASSERT now raises so the route-level tensor->layer fallback retries
layer split with the projector intact, instead of stripping --mmproj and
silently loading text-only (which returned success and skipped the fallback,
losing vision on the first load until the next cached load).
- Symmetric multi-GPU preservation: the pooled-VRAM tensor downgrade now raises
_layer_min_gpus from the usable tensor GPUs like the vision downgrade, so it
no longer collapses a multi-GPU request to a single card.
- Base the layer fallback minimum on usable GPUs: _select_gpus caps min_gpus to
the count of cards with usable VRAM, so a downgrade never forces a nearly-full
card in (or trips --fit) just to hit the count.
- Re-probe after in-app updates: key the per-binary abort cache on (path, mtime)
like _capability_cache, so POST /api/llama/update swapping the binary in place
(no backend restart) re-probes the new build instead of inheriting the old
build's abort.
- Bump _layer_min_gpus for a known-bad vision binary independent of the tensor
drop, so the route fallback's layer retry (tensor already off) still spreads
across GPUs.
Adds deterministic non-GPU regression tests for each.
* Studio: gate cached-vision layer minimum on the current tensor request
The cached-vision _layer_min_gpus bump fired for every later vision load on a
binary recorded as tensor+mmproj-incompatible, including loads that did not
request tensor parallelism. A plain non-tensor vision load that fits on one card
would then grab every GPU just because an earlier TP attempt aborted in the same
backend process.
Re-tie the bump to the current tensor request (back inside the tensor-drop
guard), so only a downgraded tensor request preserves the multi-GPU spread; a
non-tensor vision load minimizes device count as before.
* Studio: preserve GPU count + confirm assert on vision tensor fallback
Third review round on the vision/mmproj tensor-parallel fix:
- Preserve multi-GPU on the first tensor->layer fallback. The route-level retry
runs tensor-off, so the in-function downgrades can't see the original tensor
request and a fits-on-one-card model loaded the first successful fallback on a
single GPU. The GGUF load closure now passes preserve_multi_gpu_on_layer (the
toggle asked for tensor, this attempt is layer) and load_model raises
_layer_min_gpus for it, so the downgrade still spreads across GPUs.
- Cap the auto-context layer loops to usable GPUs. They bypass _select_gpus, so a
raised _layer_min_gpus could force a nearly-full card into the subset (or trip
--fit). They now start from _auto_min_gpus, capped to the GPUs with usable VRAM.
- Confirm the tensor/mmproj assert before caching. Recording (and the layer-retry
raise) now require the ggml assert marker via _is_tensor_split_assert, not the
bare-signal predicate shared with the projector-incompat branch, so a corrupt
or too-new projector that SIGSEGVs independent of split mode is no longer cached
as tensor/mmproj-incompatible.
Adds deterministic non-GPU regression tests for each.
* Studio: extend multi-GPU fallback to extra/env tensor + overhead-aware cap
Fourth review round on the vision/mmproj tensor-parallel fix:
- Preserve multi-GPU fallback for all tensor requests, not just the UI toggle.
Tensor can also be requested via --split-mode tensor in extra args or an
inherited LLAMA_ARG_SPLIT_MODE=tensor env; the fallback retries those too, so
the preserve_multi_gpu_on_layer hint now keys off _effective_tensor_parallel
(the same check the fallback uses), comparing the overall request against the
current attempt instead of only request.tensor_parallel.
- Cap the auto-context layer fallback to GPUs that can pay the per-device layer
overhead. The cap counted any card with positive usable VRAM, so a nearly-full
GPU with a few MiB free stayed eligible and could be exposed to llama.cpp and
OOM. It now mirrors _select_gpus: a card counts only if usable VRAM exceeds the
per-device pipeline overhead.
Adds deterministic non-GPU regression tests for both.
* Studio: match the #6415 split-axis assert + replay layer-preserve hint
Fifth review round on the vision/mmproj tensor-parallel fix:
- Narrow the tensor/mmproj crash signature. _is_tensor_split_assert matched any
GGML_ASSERT/GGML_ABORT, so an unrelated invariant a corrupt GGUF or projector
trips with --mmproj present could be cached as tensor/mmproj-incompatible. It
now matches the specific #6415 warmup assertion
(GGML_ASSERT(src_ss[0].axis != GGML_BACKEND_SPLIT_AXIS_0) in ggml-backend-meta),
whose split-axis signature is inherent to tensor splitting. A reworded future
assert just re-crashes-then-falls-back (vision preserved via layer split)
instead of poisoning the cache for other models.
- Persist the layer-preserve hint for respawns. A successful tensor->layer
fallback committed _last_load_kwargs without preserve_multi_gpu_on_layer, so
_respawn_if_dead replayed only --split-mode layer + tensor_parallel=False and a
mid-session respawn of a fits-on-one-card model came back single-GPU. The hint
is now in the replay snapshot, so recovery keeps the multi-GPU placement.
Adds deterministic non-GPU regression tests for both.
* Studio: tighten comments on the vision tensor-parallel fix
Make the comments and docstrings added by this PR succinct: collapse the
multi-line block comments in llama_cpp.py / inference.py to one or two lines,
trim the verbose test docstrings (the names and assert messages already carry the
intent), and shorten the module docstring. No code changes; verified comment-only
with scripts/comment_tools.py check --strip-docstrings.
* Studio: cache vision tensor abort only on the split-axis token
_is_tensor_split_assert also accepted any GGML_ASSERT/GGML_ABORT from
ggml-backend-meta, but that file holds many asserts, so an unrelated
scheduler/projector/model invariant on an --mmproj launch could cache the binary
as tensor/mmproj-incompatible and make later compatible vision models skip tensor
parallelism. Match the GGML_BACKEND_SPLIT_AXIS_* token itself (unique to the
#6415 warmup assert), not the source file name.
* Studio: don't leak the httpx test stub into later tests
The regression module stubbed httpx via sys.modules.setdefault, which installs
the lightweight stub even when real httpx is present but not yet imported. The
stub then persists for the whole pytest process, so provider/HF tests collected
later (importing httpx or huggingface_hub.errors) got a module missing
HTTPError/Response. Mirror the neighboring llama_cpp helper tests: import real
httpx first and only fall back to a stub on ImportError.
* Studio: latch the #6415 tensor-split abort on the first spawn, key it per model
The self-heal recorded the --split-mode tensor abort only in the post-retry
failure block, after the flash-attn-off retry. But SPLIT_MODE_TENSOR requires
flash_attn, so the flash-off retry can't run tensor and its output no longer
carries the warmup split-axis assert (ggml-backend-meta :541). The record
therefore never fired on the real reproducer and the crash loop repeated on
every load (reported by oobabooga on #6659).
Latch instead on the first spawn that shows the signal crash + split-axis
marker: record it, kill the process, and raise straight to the route's layer
fallback, skipping the futile flash-attn/MTP retry ladder for this crash.
The crash is a tensor-split geometry limit (e.g. MQA n_head_kv=1 splitting to
GGML_BACKEND_SPLIT_AXIS_0), not a vision/mmproj property: it reproduces without
--mmproj and even single-GPU tensor. So drop the vision/mmproj scoping, rename
_vision_tensor_* -> _tensor_split_*, and key the session cache on
(binary, mtime, model) rather than (binary, mtime) so one model's abort no
longer skips tensor for every other model on the same build.
Regression tests updated to pin the early-spawn record, the per-model cache,
and that an unrelated ggml-backend-meta assert is not treated as the marker.
* Studio: reload on explicit tensor-off after a multi-GPU layer fallback
When a tensor load is downgraded to layer but kept multi-GPU to honor the
tensor request (preserve_multi_gpu_on_layer, the geometry-cache gate, or the
budget downgrade), the server reports tensor_parallel=False with --split-mode
layer stored. A later Apply that explicitly turns the tensor toggle off then
matched the loaded state and deduped to already_loaded, so Studio kept the
fallback's all-GPU CUDA_VISIBLE_DEVICES placement instead of re-selecting
normal placement (a single GPU for a model that fits on one card).
Latch a _layer_preserves_tensor_intent flag in load_model whenever a tensor
request is downgraded to layer with the multi-GPU floor raised
(_layer_min_gpus > 1), clear it when tensor stays on or on unload, and force a
reload in _request_matches_loaded_settings when the user explicitly turns the
tensor toggle off while that flag is set. An Apply that does not touch the
toggle still dedupes, so a working multi-GPU layer server is not churned.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Studio: address reviewer.py findings on the tensor-split self-heal
P1 (dedup): tensor intent can be dropped via extras, not only the toggle. An
explicit llama_extra_args=["--split-mode", "layer"] matches the stored fallback
extras, so _request_matches_loaded_settings deduped to the preserved all-GPU
placement instead of reloading. Now reload when layer_preserves_tensor_intent
and the user explicitly drops tensor via the toggle OR via extras
(_effective_tensor_parallel of the explicit extras is false).
P1 (downgrade symmetry): the len(tp_gpus) < 2 compute-buffer downgrade cleared
tensor_parallel without raising _layer_min_gpus, unlike the budget and geometry
downgrades. GPUs below tensor's replicated compute-buffer reserve can still take
layer split's lower overhead, so keep the multi-GPU request (len(gpus) >= 2) and
let _select_gpus cap unusable cards.
P2 (cache key): key the tensor-split abort cache on st_mtime_ns, so a binary
replaced in place within the same second after an abort is re-probed instead of
inheriting the stale entry.
P2 (test hygiene): load routes/inference.py via importlib in the regression
tests instead of importing the routes package, which runs routes/__init__.py and
pulls in every router (e.g. python-multipart). Added regression coverage for the
extras-off reload, the compute-buffer multi-GPU preservation, and the same-second
nanosecond cache invalidation.
* Studio: record the tensor-split abort on the Windows CRT abort exit too
The first-spawn split-axis latch only recorded when _is_signal_crash matched
(POSIX signal or 0xC0000000+ NTSTATUS). On MSVC builds GGML_ASSERT terminates
through the CRT abort() path with exit code 3, which is neither, so the cache
never filled on Windows and every later load of the same bad binary/model
repeated the tensor crash before falling back to layer.
The split-axis marker is definitive, so accept either a signal crash or the
Windows abort() exit (3) when the marker is present. Add _is_abort_exit and a
unit test, and assert the early latch honors it.
* Studio: fix UnboundLocalError on --fit-on fallback, reload backend fast path
Two follow-ups from review on the tensor-split self-heal:
UnboundLocalError: _layer_min_gpus was initialized inside the GPU-selection try.
If NVML probing or GGUF/mmproj sizing raised, the except path logged "using
--fit on" and fell through to the command builder, where the new
self._layer_preserves_tensor_intent = _layer_min_gpus > 1 then raised, turning a
safe --fit-on layer fallback into a hard load failure. Bind _layer_min_gpus
before the try so the except path always has it.
Backend fast path: _request_matches_loaded_settings forces a reload when a
preserved tensor->layer fallback gets an explicit tensor-off request, but
load_model's own _already_in_target_state still matched the tensor-off/layer
settings and short-circuited, so the placement re-selection never ran. Mirror
the guard there: reload when layer_preserves_tensor_intent and the request drops
tensor intent. The flag clears on that reload, so there's no loop.
Added regression coverage for both.
* Studio: testable tensor-split record decision; skip futile fit-off retry
Follow-ups from a deeper review of the tensor-split self-heal:
Extract the record decision into _should_record_tensor_split_abort(rc, output)
(marker AND (signal crash OR Windows abort)) and call it from the early latch.
The combined boolean was only covered by source-inspection substring checks, so
an or->and typo would silently stop recording on Windows (CRT abort exit 3 is
not a signal) with every test still green. Add a behavioral test over the
POSIX / Windows / NTSTATUS / clean-exit / SIGKILL / no-marker matrix.
Skip the --fit off retry inside _spawn_and_wait when the crash already shows the
split-axis marker: that abort is fit-independent, so the retry just warms up and
crashes a second time before the latch records it. Skipping it lets the caller
latch immediately and corrects the latch comment.
Also clarify the dedup-guard comments (toggle read from model_fields_set vs
extras via _effective_tensor_parallel without env; the backend fast path is
intentionally broader and only ever forces a reload).
* Studio: don't reload-loop tensor-off requests under env tensor
The preserved-fallback reload guard fired on the raw tensor toggle, ignoring
LLAMA_ARG_SPLIT_MODE=tensor. For an env-driven tensor user, an explicit
tensor_parallel=false request then forced a reload that re-engaged tensor via
the env and re-created the same preserved layer fallback, so every /load
reloaded -- bypassing the env-downgrade matching that exists to avoid exactly
this loop.
Gate the guard on the env-aware effective tensor state: reload only when an
explicit toggle/extras change leaves _effective_tensor_parallel (which consults
the env) off. If the env still forces tensor, fall through to the existing
env-downgrade match, which dedupes instead of looping. Added a regression test
with LLAMA_ARG_SPLIT_MODE=tensor set.
* Studio: tighten comments and test docstrings on the TP self-heal
Condense the verbose comments and test docstrings added across the review rounds
into fewer, succinct lines without changing their intent: the early-latch and
downgrade-site rationale, the cache/key and helper docstrings, the dedup-guard
comments, and the per-test docstrings. No code changes (AST-verified comments
and docstrings only); tests and lint unchanged.
* Studio: clear preserved tensor flag on diffusion; carry it across non-drop reloads
Two follow-ups on the preserved-fallback machinery:
Diffusion: the DiffusionGemma path early-returns from load_model before the
command builder that sets/clears _layer_preserves_tensor_intent, so the flag
from a prior tensor->layer fallback leaked onto a later diffusion load and
forced needless reloads of the diffusion server on tensor-off/extra Applies.
Clear it when starting diffusion.
Settings reload: the preserve hint was recomputed only from the new request, so
a reload for an unrelated setting (e.g. max_seq_length) with the tensor toggle
omitted dropped a preserved multi-GPU layer placement back to one GPU. Carry
llama_backend.layer_preserves_tensor_intent into the hint when the request is
not an explicit tensor-off/extras-off drop, so a fitting model stays multi-GPU.
Added regression tests for the diffusion clear, the carry-forward, and the
updated tensor-intent computation.
* Studio: gate the preserve carry-forward on the same model being loaded
The tensor-intent carry-forward read llama_backend.layer_preserves_tensor_intent
without checking it belonged to the model being loaded. On a direct model switch
(load B without an explicit /unload of A), the flag is still set from A's
downgrade (it isn't reset until B's load_model reaches the command builder, after
the route reads it), so a plain load of B got preserve_multi_gpu_on_layer=True
and was spread across all GPUs even though it fits on one and the user never
requested tensor for it. The backend dedup doesn't have this leak (it checks
model_identifier first); the leak was only in the route hint.
Extract the decision into _carry_preserved_tensor_intent(preserved, same_model,
explicit_drop) and gate it on the backend still holding the same model. Add a
behavioral truth-table test (catches a `not` inversion and a missing same-model
guard) and tighten the compute-buffer downgrade test to bound its source window.
* Studio: match the HF quant too when carrying preserved tensor intent
The same-model guard on the preserve carry-forward compared only model_identifier,
which is variant-agnostic for HF repos. A later load of the same repo with a
different gguf_variant (which already bypassed dedupe on the variant mismatch)
was treated as the same model, so a request that omits tensor settings inherited
the prior variant's preserved intent and forced multi-GPU layer placement for a
quant that never requested tensor. Also require the loaded hf_variant to match for
HF repos (local direct-file loads already differ by model_identifier path). Added
a regression test for the variant guard.
* Studio: match the loaded GGUF by path too when carrying preserved tensor intent
A local directory holding multiple GGUF variants keeps one variant-agnostic
model_identifier (the directory) while config.gguf_file selects the file, so the
same-model guard let variant B inherit variant A's preserved tensor->layer
fallback and forced B onto multi-GPU. Mirror _already_in_target_state's identity
logic: match by resolved path when both sides have a local file, else by HF
variant. #6659
* Studio: let implicit same-settings reloads dedupe after a preserved fallback
The backend _already_in_target_state mirror forced a reload on ANY effective
tensor-off request once a tensor->layer fallback was preserved. In the HF
auto-pick / local-directory flows the route-level dedup is skipped, so an
identical /load with tensor omitted reached this guard and reloaded every time
even without an explicit drop. Thread the route's preserve_multi_gpu_on_layer
decision in so only an explicit drop reloads; implicit carry-forward dedupes. #6659
* Studio: only an explicit tensor/split-mode change drops preserved intent
The explicit-drop test treated request.llama_extra_args is not None as a drop,
so a same-model reload that merely added an unrelated pass-through arg (e.g.
--top-k 20) without touching the tensor field or --split-mode disabled the
carry-forward and collapsed a fitting model back to one GPU. A drop now requires
an explicit tensor_parallel field change or a non-tensor --split-mode override,
via a shared _is_explicit_tensor_drop helper used by both the already-loaded
dedup and the load carry-forward so the two readers agree. #6659
* Studio: treat an explicit clear of extras as a tensor drop
When tensor intent was extras-driven (--split-mode tensor) and fell back to a
preserved layer split, a later request that explicitly clears extras
(llama_extra_args=[]) but omits tensor_parallel left the empty list with no
split-mode override, so the carry-forward kept the model pinned multi-GPU instead
of returning to normal layer selection. _is_explicit_tensor_drop now also counts
an explicit empty-list clear as a drop, while an unrelated extra (--top-k) or
inherit (None) still carries the preserved intent. #6659
* Studio: don't treat the UI's tensor_parallel echo as a tensor drop
The Studio frontend always sends tensor_parallel and copies the /load response's
resolved value back into its state, so after a tensor->layer fallback every
ctx/settings reload carries tensor_parallel=false even though the user never
changed it. Keying the drop on the field (or on an empty extras clear) collapsed
the preserved multi-GPU placement on the next reload. A fallback also always
stores --split-mode layer, never a tensor split mode, so a clear never wipes
tensor intent. _is_explicit_tensor_drop now drops only on an explicit non-tensor
--split-mode override; the bare field echo, an empty clear, an unrelated extra,
and inherit all keep the preserved placement, and --split-mode tensor /
tensor_parallel=true re-engage tensor. #6659
* Studio: match the resolved config.identifier when carrying tensor intent
The same-model guard for the carry-forward compared the raw request id, but
ModelConfig.from_identifier normalizes it (adds the unsloth/ prefix for a
shorthand, fixes repo-id case) before load_model stores config.identifier. So a
ctx/settings reload using the shorthand id missed the match, dropped
_carry_preserved_tensor_intent, and could collapse a preserved multi-GPU layer
placement to one GPU. Compare against config.identifier (what the backend stores),
keeping it symmetric with _already_in_target_state. #6659
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Fix fast_inference crash on ABI-broken vLLM: force-load compiled extensions in the broken-vLLM probe
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Broaden broken-vLLM probe: catch non-libcudart .so failures and _moe_C_stable_libtorch
* Revert stray reformat of the PDL fix log line
* Trim verbose comments in the broken-vLLM probe
* Drop non-existent vllm._moe_C_stable_libtorch from the broken-vLLM probe
* Shorten comments in broken vLLM extension detection
Condense the docstrings and inline comments for the lazy-loaded vLLM probe
and the new regression test while keeping the rationale. Comments only, no
code changes (verified with an AST signature check and the existing tests).
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Daniel Han <danielhanchen@gmail.com>
Two intermittent Studio CI failures, both runner-environment flakes unrelated
to test logic:
Windows 'Studio install + inference without Visual Studio': the 'Hide Visual
Studio + CMake' step renames C:\Program Files\Microsoft Visual Studio to
simulate a host with no build tools. A background handle on a Program Files
directory (Defender scan or an MSBuild node) makes Rename-Item intermittently
fail with 'Access is denied', and $ErrorActionPreference = Stop turns that into
a hard job failure. Wrap the VS and cmake renames in both Hide steps in a short
Rename-WithRetry (6 tries, 3s apart) to ride out the transient lock.
macOS 'Chat UI Tests': the re-login goto to /login can be interrupted by the
SPA auth guard redirecting to the same /login URL, which Playwright reports as
'Navigation to .../login is interrupted by another navigation to .../login'.
The goto already tolerated ERR_ABORTED; broaden it to also tolerate the same-URL
interrupt (the password-field wait right after confirms we landed on /login),
and add the same signature to the two Playwright flake-retry harnesses as a
safety net for any other navigation.
Validated: playwright_chat_ui.py parses + byte-compiles, both workflow YAMLs
parse, bash -n on the retry harnesses, PowerShell AST parse on all pwsh steps,
and a functional check of Rename-WithRetry (succeeds, and rethrows after
exhausting retries).
On torch >= 2.11 torchao tries to dlopen each prebuilt _C*.so and logs a
per-file "Failed to load .../_C*.so" WARNING via the torchao logger when one
cannot load. This happens on an ABI tag mismatch in the prebuilt wheel (for
example a cp310 .so under a cp312 runtime, as on Colab) or when the kernel
targets an arch the GPU does not have (mxfp8 needs FP8 hardware, _C_cutlass_90a
is Hopper/SM90 only). torchao falls back to its non-cpp paths and Unsloth's
bnb-4bit / Triton kernels do not use these, so the warning is cosmetic.
Add a HideLoggingMessage filter on the same torchao logger that already filters
the torch < 2.11 "Skipping import of cpp extensions" message, so only these
records are dropped rather than raising the whole logger to ERROR.
Daniel Han