Skip to content

Actions: huggingface/trl

Actions

Tests (experimental)

Actions

Loading...
Loading

Show workflow options

Create status badge

Loading
tests-experimental.yml
2,500+ workflow runs
2,500+ workflow runs

Filter by Event

⌥ + click/return to exclude

Filter by Status

⌥ + click/return to exclude

Filter by Branch

⌥ + click/return to exclude

Filter by Actor

⌥ + click/return to exclude
Add an experimental GPG trainer Tests (experimental) #3948: Pull request #6681 synchronize by behroozazarkhalili
2h 2m 24s feat/3472-gpg-trainer
feat/3472-gpg-trainer
2h 2m 24s
Support sync_ref_model with PEFT by syncing the reference adapter Tests (experimental) #3947: Pull request #6975 synchronize by behroozazarkhalili
24m 10s fix/3108-sync-ref-model-peft
fix/3108-sync-ref-model-peft
24m 10s
Exclude padding from RL trainer diagnostics, and fix three related metrics Tests (experimental) #3946: Pull request #6861 synchronize by behroozazarkhalili
1h 12m 46s fix/6809-rl-metrics-masking
fix/6809-rl-metrics-masking
1h 12m 46s
Compute the loss in prediction_step so Online DPO evaluation runs Tests (experimental) #3945: Pull request #6950 synchronize by behroozazarkhalili
13m 1s fix/2228-online-dpo-eval
fix/2228-online-dpo-eval
13m 1s
test: add FSDP2 distributed coverage for AsyncGRPOTrainer Tests (experimental) #3944: Pull request #6144 synchronize by behroozazarkhalili
1h 21m 55s test/async-grpo-fsdp2-coverage
test/async-grpo-fsdp2-coverage
1h 21m 55s
Add a vllm_llm_kwargs passthrough for colocate vLLM generation Tests (experimental) #3943: Pull request #6816 synchronize by behroozazarkhalili
1h 47m 27s feat/6776-vllm-llm-kwargs
feat/6776-vllm-llm-kwargs
1h 47m 27s
Re-sync gspo_token _compute_loss with GRPOTrainer Tests (experimental) #3942: Pull request #6856 synchronize by behroozazarkhalili
1h 36m 30s fix/6807-gspo-token-resync
fix/6807-gspo-token-resync
1h 36m 30s
fix(grpo): clip the KL log-ratio before exp to avoid inf overflow Tests (experimental) #3941: Pull request #6637 synchronize by behroozazarkhalili
1h 13m 12s fix/3015-kl-exp-overflow-v2
fix/3015-kl-exp-overflow-v2
1h 13m 12s
Warn when a string model loads in float32 under mixed-precision training Tests (experimental) #3940: Pull request #6005 synchronize by behroozazarkhalili
21m 57s fix/5138-sft-fp32-bf16-warning
fix/5138-sft-fp32-bf16-warning
21m 57s
Dequantize the bitsandbytes base before the vLLM weight push Tests (experimental) #3939: Pull request #6922 synchronize by behroozazarkhalili
42m 1s fix/4973-dequantize-base-before-vllm-push
fix/4973-dequantize-base-before-vllm-push
42m 1s
Log policy_loss in GMPOTrainer and RLOOTrainer on every run Tests (experimental) #3938: Pull request #7033 synchronize by behroozazarkhalili
36m 3s fix/7005-policy-loss-metric
fix/7005-policy-loss-metric
36m 3s
Warn when reshaped sampling biases the AsyncGRPO importance ratio Tests (experimental) #3937: Pull request #6944 synchronize by behroozazarkhalili
7m 47s fix/async-grpo-truncated-logprobs
fix/async-grpo-truncated-logprobs
7m 47s
fix(minillm): compute the discounted advantage with a stable reverse recurrence Tests (experimental) #3936: Pull request #6635 synchronize by behroozazarkhalili
19m 42s fix/6626-minillm-advantage-discount
fix/6626-minillm-advantage-discount
19m 42s
fix(minillm): compute the discounted advantage with a stable reverse recurrence Tests (experimental) #3935: Pull request #6635 synchronize by behroozazarkhalili
14m 34s fix/6626-minillm-advantage-discount
fix/6626-minillm-advantage-discount
14m 34s
Re-sync gspo_token _compute_loss with GRPOTrainer Tests (experimental) #3934: Pull request #6856 synchronize by behroozazarkhalili
1h 46m 22s fix/6807-gspo-token-resync
fix/6807-gspo-token-resync
1h 46m 22s
test: add FSDP2 distributed coverage for AsyncGRPOTrainer Tests (experimental) #3933: Pull request #6144 synchronize by behroozazarkhalili
1h 38m 27s test/async-grpo-fsdp2-coverage
test/async-grpo-fsdp2-coverage
1h 38m 27s
Compute the loss in prediction_step so Online DPO evaluation runs Tests (experimental) #3932: Pull request #6950 synchronize by behroozazarkhalili
2h 33m 10s fix/2228-online-dpo-eval
fix/2228-online-dpo-eval
2h 33m 10s
Make a non-finite training step visible in log_history Tests (experimental) #3931: Pull request #6862 synchronize by behroozazarkhalili
49m 13s fix/6702-nonfinite-loss-visibility
fix/6702-nonfinite-loss-visibility
49m 13s
Exclude padding from RL trainer diagnostics, and fix three related metrics Tests (experimental) #3930: Pull request #6861 synchronize by behroozazarkhalili
1h 18m 23s fix/6809-rl-metrics-masking
fix/6809-rl-metrics-masking
1h 18m 23s
Fix the BCO encoder-decoder path Tests (experimental) #3929: Pull request #6999 synchronize by behroozazarkhalili
1h 36m 47s fix/bco-encoder-decoder
fix/bco-encoder-decoder
1h 36m 47s
Add an experimental GPG trainer Tests (experimental) #3928: Pull request #6681 synchronize by behroozazarkhalili
1h 12m 57s feat/3472-gpg-trainer
feat/3472-gpg-trainer
1h 12m 57s
Warn when reshaped sampling biases the AsyncGRPO importance ratio Tests (experimental) #3927: Pull request #6944 synchronize by behroozazarkhalili
43m 44s fix/async-grpo-truncated-logprobs
fix/async-grpo-truncated-logprobs
43m 44s
Target the Mamba in_proj in the Nemotron 3 LoRA example Tests (experimental) #3926: Pull request #6870 synchronize by behroozazarkhalili
1h 19m 16s feat/nemotron-lora-in-proj
feat/nemotron-lora-in-proj
1h 19m 16s
Dequantize the bitsandbytes base before the vLLM weight push Tests (experimental) #3925: Pull request #6922 synchronize by behroozazarkhalili
48m 24s fix/4973-dequantize-base-before-vllm-push
fix/4973-dequantize-base-before-vllm-push
48m 24s
fix(grpo): clip the KL log-ratio before exp to avoid inf overflow Tests (experimental) #3924: Pull request #6637 synchronize by behroozazarkhalili
53m 58s fix/3015-kl-exp-overflow-v2
fix/3015-kl-exp-overflow-v2
53m 58s