Hugging Face's logo

rtferraz
/
tucano2-commerce

Model card Files
xet
 Community
tucano2-commerce / notebooks

Commit History

  1. add: base vs tuned comparison cell for V4.2 final evaluation

    rtferraz
    Verified signature
    0c9199c

  1. feat(rewards): add sentiment mismatch penalty to prevent extraction reward hacking⁠

    rtferraz +1: Claude Haiku 4.5
    b1be31c

  1. fix(probe): use TRL 0.24.0 log keys — rewards/commerce_reward_fn/mean, grad_norm (not train/ prefix)

    rtferraz
    Verified signature
    080fd9a

  2. fix(classifier): reorder _classify_task_type — insights before push to prevent reengajamento misclassification

    rtferraz
    Verified signature
    63b1c86

  3. fix(rewards): 3 bugs from Cell 8 audit — push length/formal, SQL domain, extraction int check

    rtferraz
    Verified signature
    41eb15f

  4. Fix V4.2 audit: show INPUT REVIEW alongside MODEL OUTPUT for proper human scoring

    rtferraz
    Verified signature
    71422f3

  5. Fix V4.2: task weights 40/40/10/10, full audit completions, interactive input() scoring

    rtferraz
    Verified signature
    0fc9042

  6. Fix V4.2: GDPO + IWU now active in training reward path (not just monitoring)

    rtferraz
    Verified signature
    c95e44c

  7. Add V4.2 GRPO training notebook (Gold Standard, 0.5B)

    rtferraz
    Verified signature
    c5f1d2d

  1. notebooks: add V4.1 GRPO notebook (parser fix, 600 steps, LR 5e-6, constant_with_warmup)

    rtferraz
    Verified signature
    d7a090d

  1. v4: ROOT CAUSE FIX — use standard PEFT not Unsloth get_peft_model (fused LoRA kernels have dtype bug #4891). Revert to load_in_4bit=True, dtype=None matching V3.

    rtferraz
    Verified signature
    521e1d8

  2. v4: fix NF4 fp16/bf16 dtype bug (unsloth #4891) — load_in_4bit=False, 0.5B fits in full bf16 on 24GB

    rtferraz
    Verified signature
    ca397a5

  3. v4: fix fp16/bf16 mismatch — disable Unsloth gradient checkpointing (causes dtype conflict in LoRA QKV kernels at 0.5B)

    rtferraz
    Verified signature
    a40d2dc

  4. v4 notebook: fix dtype Half/BFloat16 mismatch (explicit bf16), fix tied embeddings path, fix max_length warning

    rtferraz
    Verified signature
    b1bb14c

  5. v4 notebook: fix TypeError crash, suppress warnings, update paths to CWD, add V3 task-aware system prompts

    rtferraz
    Verified signature
    631e559

  6. Fix total_mem → total_memory in V4 notebook (PyTorch API)⁠

    rtferraz +1: Claude Sonnet 4.6
    5aa00ff

  7. Add V4 Instruct-Only GRPO notebook implementing ADR-002⁠

    rtferraz +1: Claude Sonnet 4.6
    6c7b1ca

  1. apply v3 task-aware thinking controls and delete deprecated notebook⁠

    rtferraz
    1d514ac

  2. Upload grpo_vertex_v3.ipynb⁠

    rtferraz
    Verified signature
    c9b11b9

  3. Rename notebooks/grpo_vertex_v3.ipynb to notebooks/DEPRECATED_grpo_vertex_v3.ipynb⁠

    rtferraz
    Verified signature
    a62f1dc

  4. feat: add v3 notebook (.ipynb) — ready for Vertex AI Workbench

    rtferraz
    Verified signature
    6c51e5f