qgallouedec/chunked-nll-benchmark-2-bucket
2 MB
Quentin Gallouédec PRO
qgallouedec
AI & ML interests
None yet
Recent Activity
updated a Space 6 minutes ago
qgallouedec/huggingface-static-660dce published a Space 6 minutes ago
qgallouedec/huggingface-static-660dce updated a bucket 6 minutes ago
qgallouedec/huggingface-static-660dce-bucketOrganizations
Posts 3
Post
1674
TRL v1.2 introduces the SSDTrainer 🚀
Simple Self-Distillation (SSD) from Apple's paper "Embarrassingly Simple Self-Distillation Improves Code Generation" is now available as an experimental trainer in TRL.
The recipe is as minimal as the name suggests: sample completions from the model itself at a training-time temperature, then fine-tune on those raw, unverified samples with plain cross-entropy. No reward model. No verifier. No teacher model. No reinforcement learning. Just prompts and the model.
v1.2 also ships expanded tool-calling support (LLaMA 3.1 / 3.2, DeepSeek-V3), another round of KTO ↔ DPO alignment getting us closer to promoting KTO to stable, a big GRPO simplification for overlong tool results, deprecation of
Full release notes: https://github.com/huggingface/trl/releases/tag/v1.2.0
Simple Self-Distillation (SSD) from Apple's paper "Embarrassingly Simple Self-Distillation Improves Code Generation" is now available as an experimental trainer in TRL.
The recipe is as minimal as the name suggests: sample completions from the model itself at a training-time temperature, then fine-tune on those raw, unverified samples with plain cross-entropy. No reward model. No verifier. No teacher model. No reinforcement learning. Just prompts and the model.
from trl.experimental.ssd import SSDConfig, SSDTrainer
trainer = SSDTrainer(
model="Qwen/Qwen3-4B-Instruct",
args=SSDConfig(temperature=0.6, top_k=20, top_p=0.95),
train_dataset=dataset,
)
trainer.train()v1.2 also ships expanded tool-calling support (LLaMA 3.1 / 3.2, DeepSeek-V3), another round of KTO ↔ DPO alignment getting us closer to promoting KTO to stable, a big GRPO simplification for overlong tool results, deprecation of
use_transformers_paged, and key fixes for VLM response parsing.Full release notes: https://github.com/huggingface/trl/releases/tag/v1.2.0
Articles 14
Article
49
TRL v1.0: Post-Training Library Built to Move with the Field
models 789
qgallouedec/Qwen3-0.6B-SFT-20251113165959
Text Generation • 0.6B • Updated • 272
qgallouedec/tiny-aya-global-SFT
Updated
qgallouedec/tiny-aya-global-tool-calling-SFT
Updated
qgallouedec/my-other-awesome-model
Text Generation • 0.5B • Updated • 6
qgallouedec/my-awesome-model
Text Generation • 0.5B • Updated • 5
qgallouedec/trainer_output
Text Generation • 0.5B • Updated • 6
qgallouedec/test_push_output_4
Text Classification • 87.5k • Updated • 6
qgallouedec/qwen2-0.5b-deepmath-grpo
Updated
qgallouedec/my-finetuned-model
0.8B • Updated • 1
qgallouedec/Qwen3-0.6B-SFT-20251113163732
Updated
datasets 85
qgallouedec/test-grpo-vlm-log-completions
Viewer • Updated • 435 • 247
qgallouedec/llama_star_formatted
Viewer • Updated • 7.21k • 9
qgallouedec/deepmath-completions-logs2
Viewer • Updated • 48 • 50
qgallouedec/deepmath-completions-logs
Viewer • Updated • 232 • 75 • 1
qgallouedec/Dolci-Think-DPO-7B
Viewer • Updated • 150k • 12
qgallouedec/biogrid_qa
Viewer • Updated • 59.4k • 229
qgallouedec/human_gene_interaction_qa_v2
Viewer • Updated • 79.2k • 13
qgallouedec/human_gene_interaction_qa
Viewer • Updated • 1.84M • 13
qgallouedec/biogrid
Viewer • Updated • 2.82M • 1.07k
qgallouedec/trl-metrics
Viewer • Updated • 148k • 42 • 1