[STRATEGY] Refactor DeepConf strategy to use framework's step_generator#211
Open
smirnovlad wants to merge 3 commits intomainfrom
Open
[STRATEGY] Refactor DeepConf strategy to use framework's step_generator#211smirnovlad wants to merge 3 commits intomainfrom
smirnovlad wants to merge 3 commits intomainfrom
Conversation
added 3 commits
February 23, 2026 22:50
Rewrite StrategyDeepConf to use generate_step_candidates_batch() instead of direct model calls. This eliminates duplicated prompt formatting, sampling params handling, and enables FLOP tracking and checkpointing. - Replace standalone vLLM/HuggingFace/API generation with single step_generator.generate_step_candidates_batch() call - Extract logprobs from StepCandidate.other_data["raw_logprobs"] for per-token confidence computation - Support thinking mode via _complete_thinking_paths() pattern - Remove online mode (adaptive early stopping) - incompatible with step_generator pattern, can be re-added later - Remove deepconf exclusion from step_generator creation in run_tts_eval.py, add to raw vLLM list (no uncertainty wrapper) - Remove special API model handling for deepconf - Remove old experiment configs (will be recreated as needed) - Strategy reduced from 1340 to ~475 lines
Rewrite documentation to reflect the new architecture: - Remove references to direct model calls, online mode, and API-specific setup - Document step_generator-based generation flow - Update config examples to use generation section for sampling params - Add logprob format and thinking mode sections - Simplify troubleshooting
Generates traces in small batches and stops early per-sample when agreement threshold is reached, saving compute on easy problems. - Add mode/online_batch_size/min_agreement params to strategy - Extract _process_candidates and _build_result shared helpers - Refactor offline path to use shared helpers - Add _generate_online with per-sample convergence tracking - Update config, eval script, and docs
smirnovlad-test
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.