Spaces:

lebiraja
/

customer-support-env

Sleeping

App Files Files Community

customer-support-env / train

117 kB

Ctrl+K

Ctrl+K

8 contributors

History: 27 commits

lebiraja's picture

fix(train): v5 — JSON preprocessor + dual-token for friend's account

c9d0dd5 2 months ago

__init__.py

75 Bytes
feat: hackathon improvements — benchmark dashboard, GRPO training pipeline, grader fixes 3 months ago
action_parser.py

7.25 kB
fix(train): v5 — JSON preprocessor + dual-token for friend's account 2 months ago
config.py

5.84 kB
fix(train): v3 — fix fallback rewards, token budget, steps cap 2 months ago
curriculum.py

3.83 kB
feat(round2): DB-grounded multi_domain task with 30-ticket curriculum stage 3 months ago
env_client.py

3.68 kB
feat: hackathon improvements — benchmark dashboard, GRPO training pipeline, grader fixes 3 months ago
evaluate.py

5.2 kB
feat(round2): DB-grounded multi_domain task with 30-ticket curriculum stage 3 months ago
export_gguf.py

5.21 kB
fix(train): v3 — repair truncated JSON + earlier evals + fix GGUF path 2 months ago
grpo_trainer.py

7.45 kB
feat(logging): full visibility into every training step 2 months ago
local_judge.py

6.51 kB
feat(training): local judge + parallel rollout + optimized batch config 2 months ago
merge_lora.py

4.21 kB
fix: Qwen3-8B training correctness — PPO formula, think tokens, model defaults 3 months ago
model_utils.py

8 kB
fix(train): v2 — fix collapse root causes + best-ckpt tracking + GGUF export 2 months ago
prompt_builder.py

13.8 kB
fix(prompt): v4 — add explicit JSON examples for all action types 2 months ago
reward_aggregator.py

4.47 kB
fix(train): tokenize-free log-prob recompute + switch default model to Llama 3.1 2 months ago
rollout_collector.py

8.78 kB
fix(train): v3 — fix fallback rewards, token budget, steps cap 2 months ago
run_train.py

20.7 kB
fix(train): v2 — fix collapse root causes + best-ckpt tracking + GGUF export 2 months ago
sft_warmstart.py

11.7 kB
feat(train): configurable model via TRAIN_MODEL env var 3 months ago