ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Merge remote-tracking branch 'origin/main' into docker-blackwell-build

Browse source
This commit is contained in:
Daniel Han 2026-07-01 10:33:31 +00:00
commit 8fc483ec62
203 changed files with 15913 additions and 1923 deletions
Download patch file Download diff file Expand all files Collapse all files

12
.github/workflows/consolidated-tests-ci.yml vendored
View file

@ -209,7 +209,7 @@ jobs:
'peft>=0.18,<0.20' 'accelerate>=0.34,<2' \
ipython
# torchvision: unsloth_zoo.vision_utils imports it at module scope.
pip install --index-url https://download.pytorch.org/whl/cpu \
pip install --index-url https://download.pytorch.org/whl/cpu --extra-index-url https://pypi.org/simple \
'torch>=2.4,<2.11' 'torchvision<0.26'
# transformers + trl from the matrix combo.
pip install "$RESOLVED_TRANSFORMERS_SPEC"
@ -268,6 +268,10 @@ jobs:
tests/saving/test_save_shell_injection.py \
tests/saving/test_patch_saving_none_tokenizer.py \
tests/saving/test_fix_sentencepiece_gguf_robustness.py \
tests/saving/test_compressed_export_schemes.py \
tests/saving/test_export_api_surface.py \
tests/saving/test_export_dispatch.py \
tests/saving/test_imatrix_export.py \
tests/utils/test_attention_masks.py \
tests/utils/test_trunc_normal_patch.py \
tests/python/test_fast_language_model_text_only.py
@ -353,6 +357,10 @@ jobs:
tests/saving/test_save_shell_injection.py \
tests/saving/test_patch_saving_none_tokenizer.py \
tests/saving/test_fix_sentencepiece_gguf_robustness.py \
tests/saving/test_compressed_export_schemes.py \
tests/saving/test_export_api_surface.py \
tests/saving/test_export_dispatch.py \
tests/saving/test_imatrix_export.py \
tests/utils/test_attention_masks.py \
tests/utils/test_trunc_normal_patch.py \
tests/python/test_fast_language_model_text_only.py \
@ -2166,7 +2174,7 @@ jobs:
python -m pip install --upgrade pip
# Match the matrix job's torch path so unsloth_zoo's
# `import torch` resolves to the same CPU build.
pip install --index-url https://download.pytorch.org/whl/cpu \
pip install --index-url https://download.pytorch.org/whl/cpu --extra-index-url https://pypi.org/simple \
'torch>=2.4,<2.11' 'torchvision<0.26'
pip install \
'numpy<3' protobuf sentencepiece \

183
.github/workflows/mlx-ci.yml vendored
View file

@ -163,7 +163,7 @@ jobs:
'pytest==9.0.3' \
'pytest-asyncio==1.3.0' \
'httpx==0.28.1'
pip install --index-url https://download.pytorch.org/whl/cpu \
pip install --index-url https://download.pytorch.org/whl/cpu --extra-index-url https://pypi.org/simple \
'torch==2.10.0'
# github.com occasionally 500s on the git fetch; retry the
# zoo install so a single upstream blip does not fail CI.
@ -231,99 +231,6 @@ jobs:
tests/studio/test_is_mlx_dispatch_gate.py \
tests/studio/test_mlx_training_worker_behaviors.py
# Studio prebuilt llama.cpp install + GGUF inference. Mirrors the
# path Studio's setup.sh takes on macOS since #5963: plan against
# the unslothai/llama.cpp fork's latest release, which ships the
# bin-macos-arm64 bundle plus the llama-prebuilt-manifest.json the
# default policy reads. After install, downloads a small published
# GGUF (unsloth/gemma-3-270m-it-GGUF, Q4_K_M) and validates
# llama-server /completion end to end. An install failure or a
# non-zero binary exit is an Unsloth/Studio bug.
- name: Studio prebuilt llama.cpp install + GGUF inference (Mac M1)
env:
# Withheld on PR: this step runs checked-out PR code; public GGUF still downloads.
HF_TOKEN: ${{ github.event_name != 'pull_request' && secrets.HF_TOKEN || '' }}
# install_llama_prebuilt.py hits the GitHub releases API to
# resolve the asset URL. Anonymous calls share the runner-IP
# rate-limit bucket and 403 quickly -- pass the workflow's
# automatic GITHUB_TOKEN to bump us to the 5000/hr authenticated
# bucket.
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
run: |
set -euo pipefail
INSTALL_DIR="$HOME/.unsloth-studio-prebuilt-test/llama.cpp"
rm -rf "$INSTALL_DIR"
# Mirror studio/setup.sh on macOS (the install.sh user path):
# it plans against the unslothai/llama.cpp fork's latest
# release with no policy or tag flags.
python studio/install_llama_prebuilt.py \
--install-dir "$INSTALL_DIR" \
--published-repo unslothai/llama.cpp
# Studio bundles only llama-server + llama-quantize from the
# prebuilt (not llama-cli) -- inference goes through
# llama-server's HTTP /completion endpoint. Validate both:
# llama-quantize --help proves the dynamic libs link, then
# spin up llama-server and POST a /completion request on a
# tiny published GGUF.
LLAMA_SERVER="$INSTALL_DIR/build/bin/llama-server"
LLAMA_QUANT="$INSTALL_DIR/build/bin/llama-quantize"
[ -x "$LLAMA_SERVER" ] || { echo "::error::llama-server missing at $LLAMA_SERVER"; find "$INSTALL_DIR/build" -type f | head -40; exit 1; }
[ -x "$LLAMA_QUANT" ] || { echo "::error::llama-quantize missing at $LLAMA_QUANT"; exit 1; }
echo "llama-server : $LLAMA_SERVER"
echo "llama-quantize: $LLAMA_QUANT"
"$LLAMA_QUANT" --help >/dev/null && echo " llama-quantize loads OK"
mkdir -p /tmp/ggufs
bash .github/scripts/hf-download-with-retry.sh \
'unsloth/gemma-3-270m-it-GGUF' \
'gemma-3-270m-it-Q4_K_M.gguf' \
/tmp/ggufs
PORT=18080
echo "=== starting llama-server on 127.0.0.1:$PORT ==="
"$LLAMA_SERVER" \
-m /tmp/ggufs/gemma-3-270m-it-Q4_K_M.gguf \
--host 127.0.0.1 \
--port "$PORT" \
-c 256 \
-n 16 \
--no-warmup \
> /tmp/llama-server.log 2>&1 &
SERVER_PID=$!
trap 'kill "$SERVER_PID" 2>/dev/null || true' EXIT
# Wait for /health to come up
for i in $(seq 1 30); do
if curl -sf "http://127.0.0.1:$PORT/health" >/dev/null 2>&1; then
echo " server up after ${i}s"
break
fi
sleep 1
done
if ! curl -sf "http://127.0.0.1:$PORT/health" >/dev/null 2>&1; then
echo "::error::llama-server never became healthy"
tail -40 /tmp/llama-server.log
exit 1
fi
PROMPT="Hello, my name is"
echo "=== POST /completion ==="
RESP=$(curl -sf -X POST "http://127.0.0.1:$PORT/completion" \
-H 'Content-Type: application/json' \
-d "{\"prompt\":\"$PROMPT\",\"n_predict\":16,\"temperature\":0,\"seed\":3407}")
echo "raw response (head): $(echo "$RESP" | head -c 600)"
CONTENT=$(echo "$RESP" | python -c "import json,sys; print(json.loads(sys.stdin.read()).get('content',''))")
echo "completion content: $CONTENT"
if [ -z "$CONTENT" ]; then
echo "::error::llama-server /completion returned empty content"
tail -40 /tmp/llama-server.log
exit 1
fi
echo "OK: Studio prebuilt llama.cpp on Mac M1 + GGUF /completion works"
# Real MLX training + inference smoke test. Trains
# unsloth/gemma-3-270m-it for 7 deterministic LoRA steps
# (batch_size=2, gradient_accumulation_steps=3) on a single
@ -338,6 +245,9 @@ jobs:
UNSLOTH_COMPILE_DISABLE: '1'
run: |
mkdir -p mlx_workdir
# Authenticate llama.cpp's release-API lookup (anonymous 403s on rate-limit);
# read-only GITHUB_TOKEN scoped here only, never to steps that run binaries.
GH_TOKEN="${{ secrets.GITHUB_TOKEN }}" GITHUB_TOKEN="${{ secrets.GITHUB_TOKEN }}" \
python tests/studio/run_real_mlx_smoke.py train \
--workdir "$PWD/mlx_workdir"
@ -406,3 +316,88 @@ jobs:
cat "$f" 2>/dev/null || echo "(missing)"
echo
done
# Validates the macOS prebuilt path Studio's setup.sh uses (#5963): install the
# unslothai/llama.cpp fork's latest release, download a small public GGUF, and
# check llama-server /completion end to end. Split and placed last so the
# untrusted binary runs only in the final smoke step, after every HF_TOKEN step,
# leaving no token-bearing step or shared workspace for a tampered prebuilt to
# corrupt. GH_TOKEN: releases API; HF_TOKEN (withheld on PR): probe + GGUF fetch.
- name: Studio prebuilt llama.cpp install + GGUF download (Mac M1)
env:
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
HF_TOKEN: ${{ github.event_name != 'pull_request' && secrets.HF_TOKEN || '' }}
run: |
set -euo pipefail
INSTALL_DIR="$HOME/.unsloth-studio-prebuilt-test/llama.cpp"
rm -rf "$INSTALL_DIR"
# Download only -- no llama-quantize / llama-server launch in this step.
python studio/install_llama_prebuilt.py \
--install-dir "$INSTALL_DIR" \
--published-repo unslothai/llama.cpp
mkdir -p /tmp/ggufs
bash .github/scripts/hf-download-with-retry.sh \
'unsloth/gemma-3-270m-it-GGUF' \
'gemma-3-270m-it-Q4_K_M.gguf' \
/tmp/ggufs
# Final step: runs the downloaded binaries with no secrets present, and clears
# the GitHub Actions command files so a tampered prebuilt cannot influence the job.
- name: Studio prebuilt llama.cpp GGUF inference smoke (Mac M1)
run: |
set -euo pipefail
unset GITHUB_ENV GITHUB_PATH GITHUB_OUTPUT GITHUB_STEP_SUMMARY
INSTALL_DIR="$HOME/.unsloth-studio-prebuilt-test/llama.cpp"
# Studio bundles only llama-server + llama-quantize (not llama-cli);
# inference goes through llama-server's HTTP /completion endpoint.
LLAMA_SERVER="$INSTALL_DIR/build/bin/llama-server"
LLAMA_QUANT="$INSTALL_DIR/build/bin/llama-quantize"
[ -x "$LLAMA_SERVER" ] || { echo "::error::llama-server missing at $LLAMA_SERVER"; find "$INSTALL_DIR/build" -type f | head -40; exit 1; }
[ -x "$LLAMA_QUANT" ] || { echo "::error::llama-quantize missing at $LLAMA_QUANT"; exit 1; }
echo "llama-server : $LLAMA_SERVER"
echo "llama-quantize: $LLAMA_QUANT"
"$LLAMA_QUANT" --help >/dev/null && echo " llama-quantize loads OK"
PORT=18080
echo "=== starting llama-server on 127.0.0.1:$PORT ==="
"$LLAMA_SERVER" \
-m /tmp/ggufs/gemma-3-270m-it-Q4_K_M.gguf \
--host 127.0.0.1 \
--port "$PORT" \
-c 256 \
-n 16 \
--no-warmup \
> /tmp/llama-server.log 2>&1 &
SERVER_PID=$!
trap 'kill "$SERVER_PID" 2>/dev/null || true' EXIT
# Wait for /health to come up
for i in $(seq 1 30); do
if curl -sf "http://127.0.0.1:$PORT/health" >/dev/null 2>&1; then
echo " server up after ${i}s"
break
fi
sleep 1
done
if ! curl -sf "http://127.0.0.1:$PORT/health" >/dev/null 2>&1; then
echo "::error::llama-server never became healthy"
tail -40 /tmp/llama-server.log
exit 1
fi
PROMPT="Hello, my name is"
echo "=== POST /completion ==="
RESP=$(curl -sf -X POST "http://127.0.0.1:$PORT/completion" \
-H 'Content-Type: application/json' \
-d "{\"prompt\":\"$PROMPT\",\"n_predict\":16,\"temperature\":0,\"seed\":3407}")
echo "raw response (head): $(echo "$RESP" | head -c 600)"
CONTENT=$(echo "$RESP" | python -c "import json,sys; print(json.loads(sys.stdin.read()).get('content',''))")
echo "completion content: $CONTENT"
if [ -z "$CONTENT" ]; then
echo "::error::llama-server /completion returned empty content"
tail -40 /tmp/llama-server.log
exit 1
fi
echo "OK: Studio prebuilt llama.cpp on Mac M1 + GGUF /completion works"

2
.github/workflows/notebooks-ci.yml vendored
View file

@ -263,7 +263,7 @@ jobs:
# unsloth_zoo.vision_utils imports PIL at module top, and the
# easiest way to get a torch-compatible PIL on a CPU runner is
# to let torchvision pull the right Pillow version.
pip install --index-url https://download.pytorch.org/whl/cpu \
pip install --index-url https://download.pytorch.org/whl/cpu --extra-index-url https://pypi.org/simple \
'torch>=2.8,<2.11' 'torchvision<0.26'
# Pin to the same versions update_all_notebooks.py installs in
# generated notebooks. Keep these in lockstep with PIN_TRL /

4
.github/workflows/studio-backend-ci.yml vendored
View file

@ -76,7 +76,7 @@ jobs:
# Torch CPU + transformers are required by a chunk of the backend test
# suite (gpu_selection, kv_cache_estimation, utils). CPU-only torch
# keeps the install ~250 MB / ~1 min on a clean runner.
pip install --index-url https://download.pytorch.org/whl/cpu 'torch>=2.4,<2.11'
pip install --index-url https://download.pytorch.org/whl/cpu --extra-index-url https://pypi.org/simple 'torch>=2.4,<2.11'
pip install 'transformers>=4.51,<5.5'
- name: Backend tests
@ -137,7 +137,7 @@ jobs:
pyyaml jinja2 mammoth unpdf requests typer \
'numpy<3' pytest pytest-asyncio httpx
# torchvision: unsloth_zoo.vision_utils imports it at module scope.
pip install --index-url https://download.pytorch.org/whl/cpu \
pip install --index-url https://download.pytorch.org/whl/cpu --extra-index-url https://pypi.org/simple \
'torch>=2.4,<2.11' 'torchvision<0.26'
pip install 'transformers>=4.51,<5.5'
# bitsandbytes: hard import in unsloth/models/_utils.py. Recent

29
.github/workflows/studio-mac-ui-smoke.yml vendored
View file

@ -185,13 +185,14 @@ jobs:
# Retry up to 3 times to absorb known macos-14 free-runner
# flakes: (1) Playwright Node 24 pipeTransport.js 'Unexpected
# end of JSON input' crash when the Chromium browser process
# dies mid-test, and (2) Chromium net::ERR_NO_BUFFER_SPACE
# when the runner's kernel briefly runs out of socket buffers.
# The retry FULLY resets Studio (kill, reset-password, reboot,
# wait /api/health, re-export bootstrap pw) before re-running
# the script. A real test failure (assertion / timeout) does
# NOT match either pattern so it bypasses retry and surfaces
# immediately.
# dies mid-test, (2) Chromium net::ERR_NO_BUFFER_SPACE when the
# runner's kernel briefly runs out of socket buffers, and (3) a
# goto 'interrupted by another navigation' when the SPA auth
# guard redirects mid-navigation. The retry FULLY resets Studio
# (kill, reset-password, reboot, wait /api/health, re-export
# bootstrap pw) before re-running the script. A real test failure
# (assertion / timeout) does NOT match any pattern so it bypasses
# retry and surfaces immediately.
run: |
mkdir -p logs/playwright
attempt=1
@ -204,8 +205,9 @@ jobs:
if [ "$rc" -eq 0 ]; then
break
fi
if { grep -q "Unexpected end of JSON input" logs/playwright_attempt_${attempt}.log \
|| grep -q "ERR_NO_BUFFER_SPACE" logs/playwright_attempt_${attempt}.log; } \
if { grep -q "Unexpected end of JSON input" logs/playwright_attempt_${attempt}.log \
|| grep -q "ERR_NO_BUFFER_SPACE" logs/playwright_attempt_${attempt}.log \
|| grep -q "interrupted by another navigation" logs/playwright_attempt_${attempt}.log; } \
&& [ "$attempt" -lt "$max_attempts" ]; then
echo "::warning::Playwright flake on attempt ${attempt}; resetting Studio and retrying..."
kill "${STUDIO_PID}" 2>/dev/null || true
@ -280,8 +282,8 @@ jobs:
STUDIO_UI_TURN_TIMEOUT_MS: '540000'
GGUF_REPO: ${{ env.GGUF_REPO }}
GGUF_VARIANT: ${{ env.GGUF_VARIANT }}
# Same flake-retry shape as "Drive the chat UI with Playwright"
# -- catches pipeTransport JSON crash and ERR_NO_BUFFER_SPACE.
# Same flake-retry shape as "Drive the chat UI with Playwright" -- catches
# pipeTransport JSON crash, ERR_NO_BUFFER_SPACE, and nav interrupts.
run: |
mkdir -p logs/playwright_extra
attempt=1
@ -294,8 +296,9 @@ jobs:
if [ "$rc" -eq 0 ]; then
break
fi
if { grep -q "Unexpected end of JSON input" logs/playwright_extra_attempt_${attempt}.log \
|| grep -q "ERR_NO_BUFFER_SPACE" logs/playwright_extra_attempt_${attempt}.log; } \
if { grep -q "Unexpected end of JSON input" logs/playwright_extra_attempt_${attempt}.log \
|| grep -q "ERR_NO_BUFFER_SPACE" logs/playwright_extra_attempt_${attempt}.log \
|| grep -q "interrupted by another navigation" logs/playwright_extra_attempt_${attempt}.log; } \
&& [ "$attempt" -lt "$max_attempts" ]; then
echo "::warning::Playwright flake on attempt ${attempt}; resetting Studio and retrying..."
kill "${STUDIO_EXTRA_PID}" 2>/dev/null || true

24
.github/workflows/studio-windows-inference-smoke.yml vendored
View file

@ -1338,11 +1338,19 @@ jobs:
shell: pwsh
run: |
$ErrorActionPreference = 'Stop'
# A Program Files dir can hold a transient handle (Defender / MSBuild node)
# so Rename-Item intermittently fails with "Access is denied"; retry to ride it out.
function Rename-WithRetry($Path, $NewName) {
for ($i = 1; $i -le 6; $i++) {
try { Rename-Item -LiteralPath $Path -NewName $NewName -ErrorAction Stop; return }
catch { if ($i -eq 6) { throw }; Start-Sleep -Seconds 3 }
}
}
# Rename the Visual Studio install roots (incl. the Installer that holds
# vswhere.exe) so Find-VsBuildTools' vswhere + filesystem scan both miss.
foreach ($d in @("$env:ProgramFiles\Microsoft Visual Studio", "${env:ProgramFiles(x86)}\Microsoft Visual Studio")) {
if (Test-Path -LiteralPath $d) {
Rename-Item -LiteralPath $d -NewName ((Split-Path $d -Leaf) + '.vsoff')
Rename-WithRetry $d ((Split-Path $d -Leaf) + '.vsoff')
Write-Host "Hid VS: $d"
}
}
@ -1351,7 +1359,7 @@ jobs:
$hidden = @()
foreach ($c in (Get-Command cmake -All -ErrorAction SilentlyContinue)) {
if ($c.Source -and (Test-Path -LiteralPath $c.Source)) {
Rename-Item -LiteralPath $c.Source -NewName ((Split-Path $c.Source -Leaf) + '.off')
Rename-WithRetry $c.Source ((Split-Path $c.Source -Leaf) + '.off')
$hidden += $c.Source
Write-Host "Hid cmake: $($c.Source)"
}
@ -1376,7 +1384,7 @@ jobs:
- name: PyTorch CPU wheel installs and imports (no Visual Studio)
run: |
python -m pip install --upgrade pip
python -m pip install torch --index-url https://download.pytorch.org/whl/cpu
python -m pip install torch --index-url https://download.pytorch.org/whl/cpu --extra-index-url https://pypi.org/simple
python -c "import torch; print('torch', torch.__version__, 'cuda?', torch.cuda.is_available())"
- name: Install Studio (--local, --no-torch) with no build tools present
@ -1536,8 +1544,16 @@ jobs:
shell: pwsh
run: |
$ErrorActionPreference = 'Stop'
# Retry the rename: a Program Files dir can hold a transient handle that
# makes Rename-Item intermittently fail with "Access is denied".
function Rename-WithRetry($Path, $NewName) {
for ($i = 1; $i -le 6; $i++) {
try { Rename-Item -LiteralPath $Path -NewName $NewName -ErrorAction Stop; return }
catch { if ($i -eq 6) { throw }; Start-Sleep -Seconds 3 }
}
}
foreach ($d in @("$env:ProgramFiles\Microsoft Visual Studio", "${env:ProgramFiles(x86)}\Microsoft Visual Studio")) {
if (Test-Path -LiteralPath $d) { Rename-Item -LiteralPath $d -NewName ((Split-Path $d -Leaf) + '.vsoff'); Write-Host "Hid VS: $d" }
if (Test-Path -LiteralPath $d) { Rename-WithRetry $d ((Split-Path $d -Leaf) + '.vsoff'); Write-Host "Hid VS: $d" }
}
- name: Windows CUDA and ROCm prebuilts exist in unslothai/llama.cpp (what GPU users download, no VS)

2
.github/workflows/version-compat-ci.yml vendored
View file

@ -242,7 +242,7 @@ jobs:
run: |
python -m pip install --upgrade pip
# CPU torch (vllm/peft/st all depend on it).
pip install --index-url https://download.pytorch.org/whl/cpu \
pip install --index-url https://download.pytorch.org/whl/cpu --extra-index-url https://pypi.org/simple \
'torch>=2.4,<2.11' 'torchvision<0.26' 'torchcodec<0.10'
# torchcodec is a hard requirement on transformers 5.x:
# transformers/audio_utils.py:55 does