Move FP8 TE export logic to mcore.export #11409
Open
+113
−102
Conversation
This file contains 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
Signed-off-by: Piotr Kaminski <pikaminski@nvidia.com>
Signed-off-by: Laplasjan107 <Laplasjan107@users.noreply.github.com>
Laplasjan107
changed the title
Signed-off-by: Piotr Kamiński <67481570+Laplasjan107@users.noreply.github.com>
Signed-off-by: Piotr Kaminski <piotrus.kaminski@gmail.com>
Signed-off-by: Laplasjan107 <Laplasjan107@users.noreply.github.com>
Signed-off-by: Piotr Kamiński <67481570+Laplasjan107@users.noreply.github.com>
Signed-off-by: Jan Lasek <janek.lasek@gmail.com>
janekl
force-pushed
the
pikaminski/mcore_fp8_export
branch
from
ce002a8 to
71292d4
Compare
Signed-off-by: Piotr Kaminski <piotrus.kaminski@gmail.com>
Signed-off-by: Piotr Kaminski <piotrus.kaminski@gmail.com>
Signed-off-by: Laplasjan107 <Laplasjan107@users.noreply.github.com>
Signed-off-by: Piotr Kamiński <67481570+Laplasjan107@users.noreply.github.com>
janekl
reviewed
Dec 20, 2024
nemo/export/tensorrt_llm.py
Outdated
| @@ -83,6 +83,17 @@ def wrapper(*args, **kwargs): | |||
| use_pytriton = False | |||
|
|
|||
|
|
|||
| def determine_quantization_settings( | |||
There was a problem hiding this comment.
Could you add docstrings for all new / edited functions?
janekl
reviewed
Dec 20, 2024
nemo/export/tensorrt_llm.py
Outdated
| @@ -353,17 +366,38 @@ def export( | |||
| from megatron.core.export.trtllm.trtllm_helper import TRTLLMHelper | |||
| from tensorrt_llm.layers import MoeConfig | |||
|
|
|||
| use_embedding_sharing = model_configs.get("share_embeddings_and_output_weights", False) | |||
| fp8_quantized, fp8_kvcache = determine_quantization_settings( | |||
| model_configs, fp8_quantized, fp8_kvcache | |||
There was a problem hiding this comment.
It's model_configs (plural) passed, but only 1x nemo_model_config consumed in determine_quantization_settings - could we unify?
janekl
reviewed
Dec 20, 2024
| @@ -524,7 +559,7 @@ def get_transformer_config(self, nemo_model_config): | |||
| ffn_hidden_size=nemo_model_config.get('ffn_hidden_size'), | |||
| layernorm_epsilon=nemo_model_config.get('layernorm_epsilon'), | |||
| add_bias_linear=nemo_model_config.get('bias'), | |||
| num_moe_experts=nemo_model_config.get('num_moe_experts', None), | |||
| num_moe_experts=num_moe_experts if num_moe_experts > 0 else None, | |||
| normalization=transformer_config_normalization, | |||
| layernorm_zero_centered_gamma=layernorm_zero_centered_gamma, | |||
There was a problem hiding this comment.
In #11614 there is also gated_linear_unit=nemo_model_config.get('gated_linear_unit'), added.
- Can we add it here? So that we merge only this MR to offload CI slightly.
- Can we write
nemo_model_config.get('gated_linear_unit', False)to conform with the typing? There could be suprises maybe, seeNonevs0fornum_moe_experts
8 tasks
|
Please fill in the MR Usage section with details on |
janekl
reviewed
Dec 20, 2024
nemo/export/tensorrt_llm.py
Outdated
| @@ -353,17 +366,38 @@ def export( | |||
| from megatron.core.export.trtllm.trtllm_helper import TRTLLMHelper | |||
| from tensorrt_llm.layers import MoeConfig | |||
|
|
|||
| use_embedding_sharing = model_configs.get("share_embeddings_and_output_weights", False) | |||
There was a problem hiding this comment.
This overrides parameter use_embedding_sharing passed to TensorRTLLM.export(...). Should it be removed from signature?
There was a problem hiding this comment.
Also: how about keeping the original name share_embeddings_and_output_weights?
janekl
reviewed
Dec 20, 2024
| bytes_list = [state_dict[keyname][0] for keyname in keynames] | ||
| return load_scales_from_bytes(bytes_list) | ||
| decomposed_sharded_key = key.split('/') | ||
| if not len(decomposed_sharded_key): |
There was a problem hiding this comment.
Always len(decomposed_sharded_key) >= 1 so this if is inactive?
Signed-off-by: Piotr Kamiński <67481570+Laplasjan107@users.noreply.github.com>
Signed-off-by: Piotr Kaminski <piotrus.kaminski@gmail.com>
Signed-off-by: Laplasjan107 <Laplasjan107@users.noreply.github.com>
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
Signed-off-by: Piotr Kamiński <67481570+Laplasjan107@users.noreply.github.com>
|
beep boop 🤖: 🙏 The following files have warnings. In case you are familiar with these, please try helping us to improve the code base. Your code was analyzed with PyLint. The following annotations have been identified: Mitigation guide:
By applying these rules, we reduce the occurance of this message in future. Thank you for improving NeMo's documentation! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
What does this PR do ?
Support FP8 TE TRT-LLM export with mcore path
Collection: llm/nlp
Changelog
Usage
# Add a code snippet demonstrating how to use thisGitHub Actions CI
The Jenkins CI system has been replaced by GitHub Actions self-hosted runners.
The GitHub Actions CI will run automatically when the "Run CICD" label is added to the PR.
To re-run CI remove and add the label again.
To run CI on an untrusted fork, a NeMo user with write access must first click "Approve and run".
Before your PR is "Ready for review"
Pre checks:
PR Type:
If you haven't finished some of the above items you can still open "Draft" PR.
Who can review?
Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.
Additional Information