[Enhance] Simplify the pipeline of training with hf model #1461
Conversation
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
HAOCHENYE
force-pushed
the
yehc/training_with_hf
branch
4 times, most recently
from
84bbe79 to
df0fa00
Compare
hhaAndroid
reviewed
Jan 29, 2026
| self.llm_float8_handler.precompute_float8_dynamic_scale_for_fsdp(self.model.language_model) | ||
| if self.vision_float8_handler is not None and self.vision_float8_handler.enabled: | ||
|
|
||
| if self.vision_float8_handler: |
HAOCHENYE
force-pushed
the
yehc/training_with_hf
branch
4 times, most recently
from
8128974 to
b28118e
Compare
ghstack-source-id: 58a583d Pull-Request: InternLM#1449
ghstack-source-id: ea6442f Pull-Request: InternLM#1450
ghstack-source-id: d3e2426 Pull-Request: InternLM#1451
The previous `clean_param_name` only matches the "._checkpoint_wrapped_module" which starts with **.**, however, for layers wrapper with checkpoint wrapper, the layer name start with "_checkpoint_wrapped_module" cannot be cleaned for the missing prefix . ghstack-source-id: 220732d Pull-Request: InternLM#1452
…educe code duplication ghstack-source-id: bdc47bb Pull-Request: InternLM#1453
ghstack-source-id: 334484b Pull-Request: InternLM#1454
ghstack-source-id: cc6308f Pull-Request: InternLM#1455
ghstack-source-id: c0f1b93 Pull-Request: InternLM#1456
ghstack-source-id: 93b84b9 Pull-Request: InternLM#1457
`torch.autograd.grad` will raise an error if any tensor of `input` does not require gradient, e.g, the frozen `lm_head`. This commit just fix it with a simple control flow. ghstack-source-id: 2b083eb Pull-Request: InternLM#1458
ghstack-source-id: ca764b5 Pull-Request: InternLM#1460
…f16 in DecoderLayer When FSDPModule is applied to DecoderLayer, it automatically converts position_embedding to bf16, which causes the forward pass output to differ from HuggingFace models beyond the acceptable threshold. This leads to test failures for Qwen VL. The impact on text-only components is currently uncertain. ghstack-source-id: 536e8bd Pull-Request: InternLM#1462
ghstack-source-id: 943f99d Pull-Request: InternLM#1469
HAOCHENYE
force-pushed
the
yehc/training_with_hf
branch
from
b28118e to
e401b71
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.