Harden trust_remote_code consent: scan GGUF-only auto_map and drop pre-set TRC defaults (#6478)
* Scan auto_map for GGUF-only repo ids in the consent gate The trust_remote_code consent gate treated any repo classified GGUF-only (ships .gguf, no transformers-loadable weight) as having no remote code, so _config_has_auto_map returned False even when a config declared an auto_map and the repo shipped the referenced .py. The evaluator then skipped the scan/fingerprint for that target entirely. GGUF-inertness is a property of the loader, not the repo. A GGUF selection loads via llama.cpp, which never reads config.json/auto_map, and that case is already short-circuited upstream by the caller's is_gguf check (the inference route skips the remote-code preflight for a GGUF load). Every path that reaches this helper (export, training, non-GGUF inference) loads through transformers/Unsloth from_pretrained, which DOES import auto_map even for a repo that only ships .gguf weights: the custom module runs before from_pretrained fails on the missing transformers weights. The export path has no is_gguf guard and passes the source straight to FastLanguageModel.from_pretrained(trust_remote_code=True), so the in-helper GGUF skip let a repo with config.json (auto_map) + modeling_x.py + only a .gguf run unreviewed code during export. Drop the redundant repo-level GGUF short-circuit (and the now-unused _is_gguf_repo helper). A direct .gguf file reference stays inert via _is_direct_gguf_file_ref because that genuinely is a single-file llama.cpp load; repo ids are always scanned. A GGUF repo whose auto_map ships no .py still allows via the existing empty-code path, so legitimate GGUF loads are unaffected (and GGUF inference never reaches this helper at all). Only a repo that actually contains a .gguf can change behavior here; non-GGUF repos (safetensors, MLX) are byte-identical before and after. Update the GGUF auto_map test to expect a scan, and add two regression tests: a GGUF-only repo shipping auto_map Python is scanned and blocked, and a transformers-style repo (safetensors / MLX .npz) with auto_map stays scanned and blocked. * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Remove trust_remote_code config defaults; consent dialog is the only enabler trust_remote_code is a per-load decision that must go through the remote-code consent dialog, which scans the auto_map code and pins the exact version. Two pre-set paths could still enable it without the user reviewing any code, and the GGUF consent bypass rode one of them into the export flow: - 4 model_defaults YAMLs shipped trust_remote_code: true (GLM-4.7-Flash, Nemotron-3-Nano-30B-A3B, PaddleOCR-VL, ERNIE-4.5-VL). - The frontend consent hook silently enabled trust_remote_code on a clean scan whenever the caller flagged the model as needing it. Remove every trust_remote_code key from the model_defaults YAMLs (the loaders already default to False when the key is absent) and delete the frontend silent auto-enable, so trust_remote_code is only turned on after the user approves the scanned code in the dialog. The three models that genuinely run custom code ship auto_map, which the consent gate detects on its own via _config_has_auto_map, so the dialog still fires for them in inference, training, and export (Nemotron is also re-granted by the trusted-org auto-enable in the workers). GLM-4.7-Flash has no auto_map: glm4_moe_lite is native in transformers 5.0+ and it loads with trust_remote_code=False, so its YAML flag was a no-op. Adds test_yaml_trust_remote_code_removed.py. * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Drop YAML sections emptied by trust_remote_code removal Removing trust_remote_code from a model YAML whose section had no other key left a bare `inference:` header, which PyYAML parses as None; load_inference_config() then does `model_config.get("inference", {}).get(...)` and crashes on the None. Drop those now-empty section headers (24 model defaults, all the `inference:` section) so callers fall back to family/default inference params, which is the same result those models had before (their only inference override was trust_remote_code). Strengthens test_yaml_trust_remote_code_removed.py to forbid any empty/None top-level section and to load the affected models' inference config end to end. * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Add sweep asserting every model YAML loads via training + inference paths Loads all model_defaults YAMLs through load_model_defaults (training) and load_inference_config (inference) with the exact .get() access patterns the routes use, so a malformed/None section that crashes either loader is caught. * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Assert ex-TRC auto_map models still surface the consent dialog Removing the trust_remote_code YAML default must not suppress the dialog for the models that genuinely run custom code. The dialog is driven by the repo's auto_map (via preflight_remote_code_consent_for_targets -> _config_has_auto_map), not the YAML flag, so Nemotron/PaddleOCR-VL/ERNIE-4.5-VL still require consent; GLM-4.7-Flash (no auto_map) takes no dialog and loads natively. Mocks only the Hub config + .py reader. * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Tighten comments in consent-gate changes * Trim comments to be more succinct --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Daniel Han <michaelhan2050@gmail.com>
This commit is contained in:
Daniel Han
GitHub
parent
a41b8c7a44
commit
1582d2854c
77 changed files with 250 additions and 250 deletions
|
|
@ -2,7 +2,6 @@
|
|||
# Used for models without specific configurations
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -48,7 +47,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.7
|
||||
top_p: 0.95
|
||||
top_k: -1
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: unsloth/ERNIE-4.5-21B-A3B-PT
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth notebook
|
||||
|
||||
training:
|
||||
trust_remote_code: true
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -49,7 +48,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: true
|
||||
temperature: 1.5
|
||||
min_p: 0.1
|
||||
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: tiiuae/Falcon-H1-0.5B-Instruct, unsloth/Falcon-H1-0.5B-Instruct
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from Ollama
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 4096
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -45,6 +44,5 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0
|
||||
top_p: 0.9
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 4096
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -45,7 +44,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.0
|
||||
top_k: 64
|
||||
top_p: 0.95
|
||||
|
|
|
|||
|
|
@ -2,7 +2,6 @@
|
|||
# Based on Gemma2_(9B)-Alpaca.ipynb (same defaults for larger models)
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -41,6 +40,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: unsloth/gemma-2-2b-bnb-4bit, google/gemma-2-2b
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -45,7 +44,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.0
|
||||
top_k: 64
|
||||
top_p: 0.95
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -43,7 +42,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.0
|
||||
top_k: 64
|
||||
top_p: 0.95
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -43,7 +42,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.0
|
||||
top_k: 64
|
||||
top_p: 0.95
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 2
|
||||
num_epochs: 0
|
||||
|
|
@ -43,7 +42,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.0
|
||||
top_k: 64
|
||||
top_p: 0.95
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 1024
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -45,7 +44,6 @@ logging:
|
|||
audio_input: true
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.0
|
||||
top_k: 64
|
||||
top_p: 0.95
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 2
|
||||
num_epochs: 0
|
||||
|
|
@ -45,7 +44,6 @@ logging:
|
|||
audio_input: true
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.0
|
||||
top_k: 64
|
||||
top_p: 0.95
|
||||
|
|
|
|||
|
|
@ -2,7 +2,6 @@
|
|||
# Also applies to: google/gemma-4-26B-A4B-it, unsloth/gemma-4-26B-A4B-it-GGUF
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
num_epochs: 0
|
||||
learning_rate: 2e-4
|
||||
|
|
@ -40,7 +39,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.0
|
||||
top_p: 0.95
|
||||
top_k: 64
|
||||
|
|
|
|||
|
|
@ -2,7 +2,6 @@
|
|||
# Also applies to: google/gemma-4-26B-A4B
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
num_epochs: 0
|
||||
learning_rate: 2e-4
|
||||
|
|
@ -40,7 +39,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.0
|
||||
top_p: 0.95
|
||||
top_k: 64
|
||||
|
|
|
|||
|
|
@ -2,7 +2,6 @@
|
|||
# Also applies to: google/gemma-4-31B-it, unsloth/gemma-4-31B-it-GGUF
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
num_epochs: 0
|
||||
learning_rate: 2e-4
|
||||
|
|
@ -40,7 +39,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.0
|
||||
top_p: 0.95
|
||||
top_k: 64
|
||||
|
|
|
|||
|
|
@ -2,7 +2,6 @@
|
|||
# Also applies to: google/gemma-4-31B
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
num_epochs: 0
|
||||
learning_rate: 2e-4
|
||||
|
|
@ -40,7 +39,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.0
|
||||
top_p: 0.95
|
||||
top_k: 64
|
||||
|
|
|
|||
|
|
@ -2,7 +2,6 @@
|
|||
# Also applies to: google/gemma-4-E2B-it, unsloth/gemma-4-E2B-it-GGUF
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
num_epochs: 0
|
||||
learning_rate: 2e-4
|
||||
|
|
@ -40,7 +39,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.0
|
||||
top_p: 0.95
|
||||
top_k: 64
|
||||
|
|
|
|||
|
|
@ -2,7 +2,6 @@
|
|||
# Also applies to: google/gemma-4-E2B
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
num_epochs: 0
|
||||
learning_rate: 2e-4
|
||||
|
|
@ -40,7 +39,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.0
|
||||
top_p: 0.95
|
||||
top_k: 64
|
||||
|
|
|
|||
|
|
@ -2,7 +2,6 @@
|
|||
# Also applies to: google/gemma-4-E4B-it, unsloth/gemma-4-E4B-it-GGUF
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
num_epochs: 0
|
||||
learning_rate: 2e-4
|
||||
|
|
@ -40,7 +39,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.0
|
||||
top_p: 0.95
|
||||
top_k: 64
|
||||
|
|
|
|||
|
|
@ -2,7 +2,6 @@
|
|||
# Also applies to: google/gemma-4-E4B
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
num_epochs: 0
|
||||
learning_rate: 2e-4
|
||||
|
|
@ -40,7 +39,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.0
|
||||
top_p: 0.95
|
||||
top_k: 64
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 4096
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -45,7 +44,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.0
|
||||
top_p: 1.0
|
||||
top_k: 0
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 1024
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -45,7 +44,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.0
|
||||
top_p: 1.0
|
||||
top_k: 0
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -47,7 +46,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.0
|
||||
top_p: 1.0
|
||||
top_k: 0
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -47,7 +46,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.0
|
||||
top_p: 1.0
|
||||
top_k: 0
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth notebook
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -43,7 +42,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.5
|
||||
min_p: 0.1
|
||||
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: unsloth/Llama-3.2-1B-Instruct-unsloth-bnb-4bit, meta-llama/Llama-3.2-1B-Instruct, unsloth/Llama-3.2-1B-Instruct-bnb-4bit, RedHatAI/Llama-3.2-1B-Instruct-FP8, unsloth/Llama-3.2-1B-Instruct-FP8-Block, unsloth/Llama-3.2-1B-Instruct-FP8-Dynamic
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 5
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth notebook
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -45,7 +44,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.5
|
||||
min_p: 0.1
|
||||
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth notebook
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -45,7 +44,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.5
|
||||
min_p: 0.1
|
||||
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: unsloth/Meta-Llama-3.1-8B-bnb-4bit, unsloth/Meta-Llama-3.1-8B-unsloth-bnb-4bit, meta-llama/Meta-Llama-3.1-8B, unsloth/Meta-Llama-3.1-8B, unsloth/Meta-Llama-3.1-70B, meta-llama/Meta-Llama-3.1-70B, unsloth/Meta-Llama-3.1-405B-bnb-4bit, meta-llama/Meta-Llama-3.1-405B
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: "unsloth/Meta-Llama-3.1-8B-Instruct-unsloth-bnb-4bit", "meta-llama/Meta-Llama-3.1-8B-Instruct", "unsloth/Meta-Llama-3.1-8B-Instruct","RedHatAI/Llama-3.1-8B-Instruct-FP8","unsloth/Llama-3.1-8B-Instruct-FP8-Block","unsloth/Llama-3.1-8B-Instruct-FP8-Dynamic"
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 8192
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: unsloth/llama-3-8b-Instruct, meta-llama/Meta-Llama-3-8B-Instruct
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: unsloth/llama-3-8b, meta-llama/Meta-Llama-3-8B
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth notebook
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -40,7 +39,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.2
|
||||
top_p: 1.2
|
||||
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -49,7 +48,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.7
|
||||
min_p: 0.01
|
||||
top_p: 0.95
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -49,7 +48,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.15
|
||||
top_p: 0.95
|
||||
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: "unsloth/Mistral-Nemo-Base-2407", "mistralai/Mistral-Nemo-Base-2407", "unsloth/Mistral-Nemo-Instruct-2407-bnb-4bit", "unsloth/Mistral-Nemo-Instruct-2407", "mistralai/Mistral-Nemo-Instruct-2407",
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: unsloth/Mistral-Small-Instruct-2409-bnb-4bit, mistralai/Mistral-Small-Instruct-2409
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth notebook
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -43,7 +42,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.5
|
||||
min_p: 0.1
|
||||
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: unsloth/mistral-7b-instruct-v0.3, mistralai/Mistral-7B-Instruct-v0.3
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -2,7 +2,6 @@
|
|||
# Based on Mistral_v0.3_(7B)-Alpaca.ipynb
|
||||
# Also applies to: "unsloth/mistral-7b-v0.3", "mistralai/Mistral-7B-v0.3",
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -41,6 +40,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -6,7 +6,6 @@
|
|||
audio_type: dac
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
eval_steps: 0
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
|
|
@ -43,7 +42,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.4
|
||||
top_k: 40
|
||||
top_p: 0.9
|
||||
|
|
|
|||
|
|
@ -6,7 +6,6 @@
|
|||
audio_type: bicodec
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
eval_steps: 0
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
|
|
@ -48,7 +47,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.8
|
||||
top_k: 50
|
||||
top_p: 1.0
|
||||
|
|
|
|||
|
|
@ -5,7 +5,6 @@
|
|||
audio_type: csm
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
eval_steps: 0
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
|
|
@ -45,6 +44,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: unsloth/GLM-4.7-Flash-unsloth-bnb-4bit, unsloth/GLM-4.7-Flash-bnb-4bit, THUDM/GLM-4.7-Flash
|
||||
|
||||
training:
|
||||
trust_remote_code: true
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -45,7 +44,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: true
|
||||
temperature: 0.7
|
||||
top_p: 0.8
|
||||
top_k: 20
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth notebook
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -39,7 +38,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.3
|
||||
min_p: 0.15
|
||||
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: true
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -47,7 +46,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: true
|
||||
temperature: 1.0
|
||||
top_p: 1.0
|
||||
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth notebook
|
||||
|
||||
training:
|
||||
trust_remote_code: true
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -49,7 +48,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: true
|
||||
temperature: 1.5
|
||||
min_p: 0.1
|
||||
|
||||
|
|
|
|||
|
|
@ -2,7 +2,6 @@
|
|||
# Based on bert_classification.ipynb
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 1
|
||||
num_epochs: 0
|
||||
|
|
@ -41,6 +40,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -6,7 +6,6 @@
|
|||
audio_type: snac
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
eval_steps: 0
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
|
|
@ -48,7 +47,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.6
|
||||
top_p: 0.95
|
||||
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 4096
|
||||
# num_epochs: 1
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -6,7 +6,6 @@ audio_type: whisper
|
|||
audio_input: true
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
eval_steps: 5
|
||||
max_seq_length: 448
|
||||
# num_epochs: 4
|
||||
|
|
@ -41,6 +40,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: "unsloth/Phi-3-medium-4k-instruct-bnb-4bit", "microsoft/Phi-3-medium-4k-instruct",
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: "unsloth/Phi-3.5-mini-instruct-bnb-4bit", "microsoft/Phi-3.5-mini-instruct"
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -45,7 +44,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.8
|
||||
top_p: 0.95
|
||||
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# MoE model - includes gate_up_proj for MoE layers
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -46,7 +45,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.6
|
||||
top_k: 20
|
||||
top_p: 0.95
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: unsloth/Qwen2-7B-bnb-4bit, Qwen/Qwen2-7B
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth notebook
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -43,7 +42,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.5
|
||||
min_p: 0.1
|
||||
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: unsloth/Qwen2.5-1.5B-Instruct-unsloth-bnb-4bit, Qwen/Qwen2.5-1.5B-Instruct, unsloth/Qwen2.5-1.5B-Instruct-bnb-4bit
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 4096
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: unsloth/Qwen2.5-7B-unsloth-bnb-4bit, Qwen/Qwen2.5-7B, unsloth/Qwen2.5-7B-bnb-4bit
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: unsloth/Qwen2.5-Coder-1.5B-Instruct-bnb-4bit, Qwen/Qwen2.5-Coder-1.5B-Instruct
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth notebook
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -45,7 +44,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.5
|
||||
min_p: 0.1
|
||||
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
# Also applies to: unsloth/Qwen2.5-Coder-7B-Instruct, Qwen/Qwen2.5-Coder-7B-Instruct
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 32768
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -42,6 +41,3 @@ logging:
|
|||
enable_tensorboard: false
|
||||
tensorboard_dir: "runs"
|
||||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth notebook
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -43,7 +42,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 1.5
|
||||
min_p: 0.1
|
||||
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from Ollama
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 1024
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -45,7 +44,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.6
|
||||
top_k: 20
|
||||
top_p: 0.95
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from Ollama
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -45,7 +44,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.6
|
||||
top_k: 20
|
||||
top_p: 0.95
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from Ollama
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -45,7 +44,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.6
|
||||
top_k: 20
|
||||
top_p: 0.95
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# MoE model - includes gate_up_proj for MoE layers
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -46,7 +45,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.6
|
||||
top_k: 20
|
||||
top_p: 0.95
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from Ollama
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -45,7 +44,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.6
|
||||
top_k: 20
|
||||
top_p: 0.95
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -45,7 +44,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.7
|
||||
top_p: 0.80
|
||||
top_k: 20
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -45,7 +44,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.6
|
||||
top_p: 0.95
|
||||
top_k: 20
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@
|
|||
# added inference parameters from unsloth guides
|
||||
|
||||
training:
|
||||
trust_remote_code: false
|
||||
max_seq_length: 2048
|
||||
# num_epochs: 4
|
||||
num_epochs: 0
|
||||
|
|
@ -43,7 +42,6 @@ logging:
|
|||
log_frequency: 10
|
||||
|
||||
inference:
|
||||
trust_remote_code: false
|
||||
temperature: 0.7
|
||||
top_p: 0.8
|
||||
top_k: 20
|
||||
|
|
|
|||
|
|
@ -1224,9 +1224,9 @@ class TestScannerCoversAllExecutableCode:
|
|||
configs = consent._load_remote_code_configs("some/gated-repo")
|
||||
assert configs is None
|
||||
|
||||
def test_gguf_repo_auto_map_is_ignored(self):
|
||||
# A GGUF repo with a vestigial auto_map loads via llama.cpp, which never runs it,
|
||||
# so _config_has_auto_map must return False and skip the consent flow.
|
||||
def test_gguf_repo_auto_map_is_scanned_for_non_file_load_paths(self, tmp_path):
|
||||
# A GGUF-only repo id still hits export paths that run auto_map; only a direct
|
||||
# .gguf file is inert.
|
||||
def _dl(
|
||||
repo_id = None,
|
||||
filename = None,
|
||||
|
|
@ -1234,10 +1234,8 @@ class TestScannerCoversAllExecutableCode:
|
|||
**kw,
|
||||
):
|
||||
import json
|
||||
import tempfile
|
||||
|
||||
if filename == "config.json":
|
||||
p = Path(tempfile.mkdtemp()) / "config.json"
|
||||
p = tmp_path / "config.json"
|
||||
p.write_text(
|
||||
json.dumps({"auto_map": {"AutoModelForCausalLM": "modeling_decilm.X"}})
|
||||
)
|
||||
|
|
@ -1251,7 +1249,79 @@ class TestScannerCoversAllExecutableCode:
|
|||
return_value = ["config.json", "model-00001-of-00097.gguf"],
|
||||
),
|
||||
):
|
||||
assert consent._config_has_auto_map("unsloth/Some-Model-GGUF") is False
|
||||
assert consent._config_has_auto_map("unsloth/Some-Model-GGUF") is True
|
||||
|
||||
def test_gguf_only_repo_with_python_is_scanned_and_blocked(self, tmp_path):
|
||||
# Regression: the GGUF-only short-circuit must not skip auto_map Python for export loaders.
|
||||
def _dl(
|
||||
repo_id = None,
|
||||
filename = None,
|
||||
token = None,
|
||||
**kw,
|
||||
):
|
||||
import json
|
||||
|
||||
p = tmp_path / filename
|
||||
if filename == "config.json":
|
||||
p.write_text(json.dumps({"auto_map": {"AutoModel": "modeling_evil.X"}}))
|
||||
return str(p)
|
||||
if filename == "modeling_evil.py":
|
||||
p.write_text("import subprocess\nsubprocess.Popen(['id'])\n")
|
||||
return str(p)
|
||||
raise EntryNotFoundError(filename)
|
||||
|
||||
with (
|
||||
patch("huggingface_hub.hf_hub_download", side_effect = _dl),
|
||||
patch(
|
||||
"huggingface_hub.list_repo_files",
|
||||
return_value = ["config.json", "modeling_evil.py", "model.Q4_K_M.gguf"],
|
||||
),
|
||||
):
|
||||
d = evaluate_remote_code_consent_for_targets(
|
||||
["evil/GGUF-Only"],
|
||||
trust_remote_code = True,
|
||||
)
|
||||
|
||||
assert d.has_remote_code is True
|
||||
assert d.blocked is True
|
||||
assert d.max_severity == HIGH
|
||||
assert d.fingerprint
|
||||
|
||||
def test_transformers_style_repo_auto_map_is_scanned_and_blocked(self, tmp_path):
|
||||
# A non-GGUF repo (safetensors/MLX) with auto_map is still scanned and blocked.
|
||||
def _dl(
|
||||
repo_id = None,
|
||||
filename = None,
|
||||
token = None,
|
||||
**kw,
|
||||
):
|
||||
import json
|
||||
|
||||
p = tmp_path / filename
|
||||
if filename == "config.json":
|
||||
p.write_text(json.dumps({"auto_map": {"AutoModel": "modeling_evil.X"}}))
|
||||
return str(p)
|
||||
if filename == "modeling_evil.py":
|
||||
p.write_text("import subprocess\nsubprocess.Popen(['id'])\n")
|
||||
return str(p)
|
||||
raise EntryNotFoundError(filename)
|
||||
|
||||
for weights in (["model.safetensors"], ["weights.npz"]):
|
||||
with (
|
||||
patch("huggingface_hub.hf_hub_download", side_effect = _dl),
|
||||
patch(
|
||||
"huggingface_hub.list_repo_files",
|
||||
return_value = ["config.json", "modeling_evil.py", *weights],
|
||||
),
|
||||
):
|
||||
d = evaluate_remote_code_consent_for_targets(
|
||||
["org/Transformers-Style"],
|
||||
trust_remote_code = True,
|
||||
)
|
||||
assert d.has_remote_code is True, weights
|
||||
assert d.blocked is True, weights
|
||||
assert d.max_severity == HIGH, weights
|
||||
assert d.fingerprint, weights
|
||||
|
||||
def test_direct_gguf_file_reference_has_no_auto_map(self):
|
||||
# A direct .gguf file reference (repo id + filename, >=3 segments) is a GGUF load: no remote code, no Hub call.
|
||||
|
|
|
|||
163
studio/backend/tests/test_yaml_trust_remote_code_removed.py
Normal file
163
studio/backend/tests/test_yaml_trust_remote_code_removed.py
Normal file
|
|
@ -0,0 +1,163 @@
|
|||
# SPDX-License-Identifier: AGPL-3.0-only
|
||||
# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
|
||||
|
||||
"""Regression: model-default YAMLs must not pre-set trust_remote_code.
|
||||
|
||||
It is a per-load decision made through the consent dialog (which scans and pins the
|
||||
auto_map code), never a config default -- a YAML flag would re-open the no-review
|
||||
bypass. Models that run custom code ship auto_map, so the dialog still fires without it.
|
||||
"""
|
||||
|
||||
from pathlib import Path
|
||||
|
||||
import yaml
|
||||
|
||||
_CONFIGS = Path(__file__).resolve().parent.parent / "assets" / "configs"
|
||||
_MODEL_DEFAULTS = _CONFIGS / "model_defaults"
|
||||
|
||||
|
||||
def test_no_model_default_yaml_sets_trust_remote_code():
|
||||
offenders = []
|
||||
for f in _MODEL_DEFAULTS.rglob("*.yaml"):
|
||||
doc = yaml.safe_load(f.read_text()) or {}
|
||||
if not isinstance(doc, dict):
|
||||
continue
|
||||
for section, body in doc.items():
|
||||
if isinstance(body, dict) and "trust_remote_code" in body:
|
||||
offenders.append(
|
||||
f"{f.relative_to(_CONFIGS)} [{section}={body['trust_remote_code']}]"
|
||||
)
|
||||
assert not offenders, (
|
||||
"trust_remote_code must not be pre-set in model defaults; it is enabled only via "
|
||||
f"the consent dialog. Remove it from: {offenders}"
|
||||
)
|
||||
|
||||
|
||||
def test_no_model_default_yaml_has_empty_or_none_section():
|
||||
# A bare `inference:` header (no keys) parses to None and crashes the .get() loaders.
|
||||
offenders = []
|
||||
for f in _MODEL_DEFAULTS.rglob("*.yaml"):
|
||||
doc = yaml.safe_load(f.read_text())
|
||||
if not isinstance(doc, dict):
|
||||
offenders.append(f"{f.relative_to(_CONFIGS)} (not a mapping)")
|
||||
continue
|
||||
for section, body in doc.items():
|
||||
if body is None or (isinstance(body, dict) and not body):
|
||||
offenders.append(f"{f.relative_to(_CONFIGS)} [{section}]")
|
||||
assert not offenders, (
|
||||
"empty/None YAML section would crash the config loaders; drop the bare section "
|
||||
f"header instead. Offending: {offenders}"
|
||||
)
|
||||
|
||||
|
||||
def test_formerly_flagged_models_load_inference_config_without_crash():
|
||||
# Models whose inference section was emptied by the TRC removal must still load.
|
||||
from utils.inference import load_inference_config
|
||||
for model in (
|
||||
"tiiuae/Falcon-H1-0.5B-Instruct",
|
||||
"unsloth/Llama-3.2-1B-Instruct",
|
||||
"unsloth/Qwen2.5-7B",
|
||||
):
|
||||
cfg = load_inference_config(model)
|
||||
assert isinstance(cfg, dict)
|
||||
assert cfg.get("trust_remote_code", False) is False
|
||||
|
||||
|
||||
def test_all_model_yamls_load_for_training_and_inference():
|
||||
# Every YAML must load through both config paths (training + inference) as the routes do.
|
||||
from utils.inference import load_inference_config
|
||||
from utils.models.model_config import load_model_defaults
|
||||
|
||||
infer_keys = {
|
||||
"temperature",
|
||||
"top_p",
|
||||
"top_k",
|
||||
"min_p",
|
||||
"presence_penalty",
|
||||
"trust_remote_code",
|
||||
}
|
||||
failures = []
|
||||
for f in sorted(_MODEL_DEFAULTS.rglob("*.yaml")):
|
||||
stem = f.stem
|
||||
try:
|
||||
md = load_model_defaults(stem)
|
||||
assert isinstance(md, dict), f"load_model_defaults -> {type(md).__name__}"
|
||||
assert not [k for k, v in md.items() if v is None], "has a None section"
|
||||
# the dict sections the loaders read via .get('sect', {}).get(...)
|
||||
for sect in ("training", "inference", "lora", "logging"):
|
||||
assert isinstance(md.get(sect, {}), dict), f"{sect!r} is not a mapping"
|
||||
md.get("training", {}).get("trust_remote_code", False) # routes/training.py:263
|
||||
cfg = load_inference_config(stem)
|
||||
assert infer_keys <= set(cfg), f"inference config missing {infer_keys - set(cfg)}"
|
||||
except Exception as e: # noqa: BLE001 - aggregate so one failure does not hide others
|
||||
failures.append(f"{f.relative_to(_CONFIGS)}: {type(e).__name__}: {e}")
|
||||
assert not failures, "YAML config loaders crashed on: " + "; ".join(failures)
|
||||
|
||||
|
||||
def test_base_templates_have_no_trust_remote_code():
|
||||
for name in ("full_finetune.yaml", "lora_text.yaml", "vision_lora.yaml"):
|
||||
doc = yaml.safe_load((_CONFIGS / name).read_text()) or {}
|
||||
flat = yaml.safe_dump(doc)
|
||||
assert "trust_remote_code" not in flat, f"{name} should not set trust_remote_code"
|
||||
|
||||
|
||||
def test_loader_defaults_trust_remote_code_off_for_formerly_flagged_models():
|
||||
# The 4 models that used to ship trust_remote_code: true must now report no default.
|
||||
from utils.models.model_config import load_model_defaults
|
||||
for model in (
|
||||
"unsloth/GLM-4.7-Flash",
|
||||
"unsloth/Nemotron-3-Nano-30B-A3B",
|
||||
"unsloth/PaddleOCR-VL",
|
||||
"unsloth/ERNIE-4.5-VL-28B-A3B-PT",
|
||||
):
|
||||
d = load_model_defaults(model)
|
||||
for section in ("training", "inference"):
|
||||
assert not (d.get(section) or {}).get(
|
||||
"trust_remote_code", False
|
||||
), f"{model} [{section}] still carries a trust_remote_code default"
|
||||
|
||||
|
||||
def test_formerly_flagged_auto_map_models_still_require_consent_dialog():
|
||||
# Crux: an auto_map model must STILL surface the dialog (driven by auto_map, not the
|
||||
# YAML flag). Real backend path, mocking only the Hub json + .py fetch.
|
||||
from unittest.mock import patch
|
||||
from utils.security import consent, preflight_remote_code_consent_for_targets
|
||||
|
||||
auto_map_cfg = [
|
||||
{
|
||||
"auto_map": {
|
||||
"AutoConfig": "configuration_x.XConfig",
|
||||
"AutoModelForCausalLM": "modeling_x.XForCausalLM",
|
||||
}
|
||||
}
|
||||
]
|
||||
benign_py = {"modeling_x.py": "class XForCausalLM:\n pass\n"}
|
||||
for model in (
|
||||
"unsloth/Nemotron-3-Nano-30B-A3B",
|
||||
"unsloth/PaddleOCR-VL",
|
||||
"unsloth/ERNIE-4.5-VL-28B-A3B-PT",
|
||||
):
|
||||
with (
|
||||
patch.object(consent, "_load_remote_code_configs", return_value = auto_map_cfg),
|
||||
patch.object(consent, "repo_remote_code_files", return_value = benign_py),
|
||||
):
|
||||
decision = preflight_remote_code_consent_for_targets([model], hf_token = None)
|
||||
# routes/models.py opens the dialog from decision.has_remote_code.
|
||||
assert decision.has_remote_code is True, (
|
||||
f"{model} ships auto_map but the consent scan did not flag it -> dialog would "
|
||||
"not fire"
|
||||
)
|
||||
|
||||
|
||||
def test_no_auto_map_model_takes_no_dialog():
|
||||
# Flip side: GLM-4.7-Flash ships no auto_map -> no dialog; its old YAML flag was a no-op.
|
||||
from unittest.mock import patch
|
||||
from utils.security import consent, preflight_remote_code_consent_for_targets
|
||||
|
||||
with patch.object(
|
||||
consent, "_load_remote_code_configs", return_value = [{"model_type": "glm4_moe_lite"}]
|
||||
):
|
||||
decision = preflight_remote_code_consent_for_targets(
|
||||
["unsloth/GLM-4.7-Flash"], hf_token = None
|
||||
)
|
||||
assert decision.has_remote_code is False
|
||||
|
|
@ -85,8 +85,13 @@ def _config_has_auto_map(model_name: str, hf_token: Optional[str] = None) -> Opt
|
|||
"""Whether any config (model/tokenizer/processor) declares an ``auto_map`` the load
|
||||
would execute. Reads raw JSON with ``hf_token``; returns None when a config is
|
||||
unreadable (transient/auth) so the caller treats it as "unknown" and scans, False
|
||||
when the repo genuinely ships none. GGUF is False (llama.cpp never runs auto_map);
|
||||
this is the single chokepoint for that rule, shared by validate / scan / worker.
|
||||
when the repo genuinely ships none.
|
||||
|
||||
GGUF-inertness is the LOADER's property, decided upstream by the caller's ``is_gguf``
|
||||
check, not here. Every path that reaches this helper (export, training, non-GGUF
|
||||
inference) loads via ``from_pretrained``, which imports ``auto_map`` even for a
|
||||
``.gguf``-only repo, so a GGUF-classified repo id MUST still be scanned. Only a direct
|
||||
``.gguf`` FILE reference is inert (a genuine single-file llama.cpp load).
|
||||
"""
|
||||
# A direct .gguf FILE loads via llama.cpp (auto_map inert). A bare repo id ending in
|
||||
# .gguf can still ship safetensors + auto_map, so it falls through to the scan.
|
||||
|
|
@ -97,11 +102,6 @@ def _config_has_auto_map(model_name: str, hf_token: Optional[str] = None) -> Opt
|
|||
return None
|
||||
if not any(bool((cfg or {}).get("auto_map")) for cfg in configs):
|
||||
return False
|
||||
# auto_map present but a GGUF repo -> inert. Checked only when auto_map exists, so
|
||||
# normal models skip the extra listing.
|
||||
if _is_gguf_repo(model_name, hf_token):
|
||||
logger.debug("Ignoring auto_map for GGUF repo '%s' (llama.cpp never runs it).", model_name)
|
||||
return False
|
||||
return True
|
||||
|
||||
|
||||
|
|
@ -124,42 +124,6 @@ def _is_direct_gguf_file_ref(model_name: str) -> bool:
|
|||
return name.count("/") >= 2
|
||||
|
||||
|
||||
# Weight formats transformers can load (and thus run auto_map for). A repo shipping any
|
||||
# of these is not GGUF-only -- the user could load it through transformers -- so consent
|
||||
# still applies even if it also ships a .gguf.
|
||||
_TRANSFORMERS_WEIGHT_SUFFIXES = (
|
||||
".safetensors",
|
||||
".bin",
|
||||
".pt",
|
||||
".pth",
|
||||
".h5",
|
||||
".msgpack",
|
||||
".onnx",
|
||||
".ckpt",
|
||||
)
|
||||
|
||||
|
||||
def _is_gguf_repo(model_name: str, hf_token: Optional[str] = None) -> bool:
|
||||
"""Whether a remote repo loads only through llama.cpp (GGUF weights and NO
|
||||
transformers-loadable weights), making its config inert. A repo that also ships
|
||||
transformers weights is NOT GGUF (auto_map could run, so still gate). A listing
|
||||
failure is treated as "not known-GGUF" (fall through to scan).
|
||||
"""
|
||||
try:
|
||||
from utils.paths import is_local_path
|
||||
|
||||
if is_local_path(model_name):
|
||||
return False
|
||||
from huggingface_hub import list_repo_files
|
||||
|
||||
files = [f.lower() for f in list_repo_files(model_name, token = hf_token)]
|
||||
has_gguf = any(f.endswith(".gguf") for f in files)
|
||||
has_transformers_weights = any(f.endswith(_TRANSFORMERS_WEIGHT_SUFFIXES) for f in files)
|
||||
return has_gguf and not has_transformers_weights
|
||||
except Exception:
|
||||
return False
|
||||
|
||||
|
||||
def _load_remote_code_configs(model_name: str, hf_token: Optional[str] = None) -> Optional[list]:
|
||||
"""Read every config that can declare ``auto_map`` (model/tokenizer/processor) as
|
||||
raw dicts. Returns the configs present (``[]`` when all 404, a definitive "no
|
||||
|
|
@ -300,8 +264,7 @@ def evaluate_remote_code_consent_for_targets(
|
|||
)
|
||||
|
||||
if not combined:
|
||||
# auto_map declared but no executable .py (e.g. a GGUF repo's vestigial
|
||||
# auto_map) -> nothing to run -> allow.
|
||||
# auto_map declared but no executable .py (e.g. GGUF repo) -> nothing to scan -> allow.
|
||||
return RemoteCodeDecision(
|
||||
primary,
|
||||
False,
|
||||
|
|
|
|||
|
|
@ -45,11 +45,9 @@ export async function confirmRemoteCodeIfNeeded({
|
|||
};
|
||||
}
|
||||
|
||||
// Open the dialog for custom-code consent OR flagged unsafe files. Otherwise a model
|
||||
// can still need trust_remote_code via its YAML default (no auto_map, e.g. GLM-4.7-Flash):
|
||||
// propagate the caller's requirement with an empty pin instead of sending false.
|
||||
// No custom code and nothing unsafe: proceed without trust_remote_code. Models needing
|
||||
// it ship auto_map and hit the dialog below, so the flag is only enabled via approval.
|
||||
if (!scan.requiresTrustRemoteCode && scan.unsafeFiles.length === 0) {
|
||||
if (requiresTrustRemoteCode) onApprove(null);
|
||||
return true;
|
||||
}
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue