[Feature] GLM-45-AIR Support Mix Quantization(Dense wfp8afp8 and wint8 triton_moe_backend) #4051
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
|
Thanks for your contribution! |
ckl117
changed the title
YuanRisheng
reviewed
Sep 10, 2025
fastdeploy/worker/worker_process.py
Outdated
Comment on lines
730
to
734
| if quant_config_name == "wint8" and "Glm4Moe" in model_config.architectures[0]: | ||
| quantization_config["dense_quant_type"] = "wfp8afp8" | ||
| quantization_config["moe_quant_type"] = "wint8" | ||
| quantization_config["quantization"] = "mix_quant" | ||
| quant_config_name = "mix_quant" |
Collaborator
There was a problem hiding this comment.
这个不能放到config.json里进行配置吗，前边ernie那个是特殊情况才这样写的
Collaborator
Author
There was a problem hiding this comment.
我理解config.json应该不能随意改吧，config里的quant配置应该和模型权重强相关。GLM是因为shape原因，dense无法走wint8。
合理的方法应该通过服务启动参数 --quantization 指定一个dict完成mix在线量化配置。
ckl117
changed the title
yuanlehome
reviewed
Sep 11, 2025
Comment on lines
+183
to
+184
| if not self.quant_config.is_checkpoint_bf16: | ||
| return |
Collaborator
There was a problem hiding this comment.
这个直接reture是合理的吗？会存在这种情况吗？
Collaborator
Author
There was a problem hiding this comment.
cutlass的weight_only也加了这个判断。当前wint8都是在线量化，应该不会走到这个分支。如果有离线量化好的权重应该会走到这个分支
yuanlehome
previously approved these changes
Sep 11, 2025
YuanRisheng
approved these changes
Sep 11, 2025
ckl117
added a commit
to ckl117/FastDeploy
that referenced
this pull request
Sep 11, 2025
…8 triton_moe_backend) (PaddlePaddle#4051)
qingqing01
pushed a commit
that referenced
this pull request
Sep 15, 2025
* [Feature] Support zai-org/GLM-4.5-Air BF16 model (#3928) * support glm45_air * [Feature] GLM-45-AIR Support Mix Quantization(Dense wfp8afp8 and wint8 triton_moe_backend) (#4051) * check * fix v1 load for mix and wint8 * check --quantizations 'None' * check * support RL rollout * check v1 loader * check glm rollout_model, change wfp8afp8 per_token_cast_to_fp8 to native impl * check rollout moe gate begin layer_id * check rollout e_score_correction_bias * delete infer_to_train_mapping={} * code check
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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.
GLM-45-AIR Support wint8
Remove duplicate code about noaux_tc op.
Support WFP8AFP8LinearMethod and TritonWeightOnlyMoEMethod v1 loader