Fix weight tying logic between _tied_weights_keys and tie_word_embeddings
#42385
Conversation
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
|
run-slow: mistral |
|
This comment contains models: ["models/mistral"] |
CI Results✅ No failing test specific to this PR 🎉 ! |
|
run-slow: mistral3 |
|
This comment contains models: ["models/mistral3"] |
CI ResultsModel CI Report❌ Failed tests
|
|
run-slow: mistral, mistral3 |
|
This comment contains models: ["models/mistral", "models/mistral3"] |
CI ResultsModel CI Report❌ Failed tests
|
|
run-slow: olmoe |
|
This comment contains models: ["models/olmoe"] |
CI Results✅ No failing test specific to this PR 🎉 ! |
|
run-slow: olmoe, mistral, mistral3 |
|
This comment contains models: ["models/mistral", "models/mistral3", "models/olmoe"] |
CI Results✅ No failing test specific to this PR 🎉 ! |
|
run-slow: olmoe, mistral, mistral3, phi3, starcoder |
|
This comment contains models: ["models/mistral", "models/mistral3", "models/olmoe", "models/phi3"] |
|
run-slow: olmoe, mistral, mistral3, phi3, starcoder, bigbird, bigbird_pegasus |
CI Results |
|
run-slow: olmoe, mistral, mistral3, phi3, starcoder, bigbird, bigbird_pegasus, whisper, internvl, llava, llava_next, llava_next_video, qwen, fsmt |
|
This comment contains models: ["models/bigbird_pegasus", "models/fsmt", "models/internvl", "models/llava", "models/llava_next", "models/llava_next_video", "models/mistral", "models/mistral3", "models/olmoe", "models/phi3", "models/whisper"] |
CI Results✅ No failing test specific to this PR 🎉 ! |
|
run-slow: olmoe, mistral, mistral3, phi3, starcoder, bigbird, bigbird_pegasus, whisper, internvl, llava, llava_next, llava_next_video, qwen, fsmt, video_llava, deepseek_v3 |
|
This comment contains models: ["models/bigbird_pegasus", "models/deepseek_v3", "models/fsmt", "models/internvl", "models/llava", "models/llava_next", "models/llava_next_video", "models/mistral", "models/mistral3", "models/olmoe", "models/phi3", "models/video_llava", "models/whisper"] |
CI Results✅ No failing test specific to this PR 🎉 ! |
|
[For maintainers] Suggested jobs to run (before merge) run-slow: fsmt |
|
run-slow: olmoe, mistral, mistral3, phi3, starcoder, bigbird, bigbird_pegasus, whisper, internvl, llava, llava_next, llava_next_video, qwen, fsmt, video_llava, deepseek_v3, qwen3_vl_moe |
|
This comment contains models: ["models/bigbird_pegasus", "models/deepseek_v3", "models/fsmt", "models/internvl", "models/llava", "models/llava_next", "models/llava_next_video", "models/mistral", "models/mistral3", "models/olmoe", "models/phi3", "models/qwen3_vl_moe", "models/video_llava", "models/whisper"] |
CI Results✅ No failing test specific to this PR 🎉 ! |
molbap
changed the title
_tied_weights_keys and tie_word_embeddings
|
[For maintainers] Suggested jobs to run (before merge) run-slow: fsmt |
|
I removed the fallback to parent config in case |
|
A remaining headscratcher is should_tie = tie_encoder_decoder if tie_word_embeddings is None else tie_word_embeddingsand it seems to work/keep the key as source of authority. LMK! |
|
[For maintainers] Suggested jobs to run (before merge) run-slow: fsmt, kyutai_speech_to_text, musicgen, musicgen_melody |
1 similar comment
|
[For maintainers] Suggested jobs to run (before merge) run-slow: fsmt, kyutai_speech_to_text, musicgen, musicgen_melody |
What does this PR do?