Skip to content

Add GPT-OSS converter with heterogeneous block pattern support #374

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 17 commits into
base: main
Choose a base branch
from
Open

Add GPT-OSS converter with heterogeneous block pattern support #374

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
17 commits
Select commit Hold shift + click to select a range
265b23b
Add GPT-OSS converter with heterogeneous block pattern support
tscholak Oct 12, 2025
d72a444
Fix GPT-OSS converter bugs and disable Triton sparse_map kernel
tscholak Oct 14, 2025
7dcf0dd
Fix architecture/architectures field typo in HuggingFace checkpoint h…
tscholak Oct 14, 2025
4cadbbd
Refactor GPT-OSS converter based on PR review feedback
tscholak Oct 14, 2025
1440e65
Fix YARN RoPE export and import in Llama converter
tscholak Oct 14, 2025
bd309cd
Add pattern cycle detection for GPT-OSS layer_types import
tscholak Oct 14, 2025
400e422
Fix GPT-OSS converter for MXFP4 quantized checkpoints
tscholak Oct 15, 2025
3d55283
WIP: Add MoE bias support for GPT-OSS converter
tscholak Oct 15, 2025
9a77d7c
Add MoE bias support for GPT-OSS converter
tscholak Oct 15, 2025
69d2393
Fix MoE weight dimensions using transposed_weight parameter
tscholak Oct 15, 2025
ddfde72
Add MoE weight converter for GPT-OSS format
tscholak Oct 15, 2025
9251818
Add GPT-OSS forward pass validation and custom GLU activation
tscholak Oct 16, 2025
52e675b
WIP: Add debugging traces and test scripts for GPT-OSS MoE conversion
tscholak Oct 16, 2025
bfceea3
Fix looped MoE transpose bug and add FSDP weight access documentation
tscholak Oct 21, 2025
803097c
Fix GPT-OSS MoE weight format and looped implementation
tscholak Oct 21, 2025
1757988
wip
tscholak Oct 21, 2025
93dbfc8
Fix Mixtral MoE layer configuration to use moe_affine_linear
tscholak Oct 24, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions fast_llm/models/gpt/config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,7 @@
AutoGPTHuggingfaceCheckpointFormat,
DiffusionDreamCheckpointFormat,
DiffusionLlamaCheckpointFormat,
GptOssCheckpointFormat,
LlamaCheckpointFormat,
MistralCheckpointFormat,
MixtralCheckpointFormat,
Expand Down Expand Up @@ -117,6 +118,7 @@ class GPTModelConfig(FastLLMModelConfig):
DiffusionDreamCheckpointFormat,
DiffusionLlamaCheckpointFormat,
AprielHybridSSMCheckpointFormat,
GptOssCheckpointFormat,
)

@classmethod
Expand Down
3 changes: 3 additions & 0 deletions fast_llm/models/gpt/conversion/auto.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,7 @@
AprielHybridSSMCheckpointFormat,
DiffusionDreamCheckpointFormat,
DiffusionLlamaCheckpointFormat,
GptOssCheckpointFormat,
LlamaCheckpointFormat,
MistralCheckpointFormat,
MixtralCheckpointFormat,
Expand All @@ -15,6 +16,7 @@
)
from fast_llm.models.gpt.conversion.diffusion_dream import DiffusionDreamHuggingfaceCheckpointHandler
from fast_llm.models.gpt.conversion.diffusion_llama import DiffusionLlamaHuggingfaceCheckpointHandler
from fast_llm.models.gpt.conversion.gpt_oss import GptOssHuggingfaceCheckpointHandler
from fast_llm.models.gpt.conversion.llama import LlamaHuggingfaceCheckpointHandler
from fast_llm.models.gpt.conversion.mistral import MistralHuggingfaceCheckpointHandler
from fast_llm.models.gpt.conversion.mixtral import MixtralHuggingfaceCheckpointHandler
Expand All @@ -35,4 +37,5 @@ class AutoGPTHuggingfaceCheckpointHandler(
DiffusionDreamCheckpointFormat.name: DiffusionDreamHuggingfaceCheckpointHandler,
DiffusionLlamaCheckpointFormat.name: DiffusionLlamaHuggingfaceCheckpointHandler,
AprielHybridSSMCheckpointFormat.name: AprielHuggingfaceCheckpointHandler,
GptOssCheckpointFormat.name: GptOssHuggingfaceCheckpointHandler,
}
4 changes: 4 additions & 0 deletions fast_llm/models/gpt/conversion/config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,3 +47,7 @@ class DiffusionLlamaCheckpointFormat(GPTHuggingfaceCheckpointFormat):

class AprielHybridSSMCheckpointFormat(GPTHuggingfaceCheckpointFormat):
name: typing.ClassVar[str] = "apriel_hybrid_ssm"


class GptOssCheckpointFormat(GPTHuggingfaceCheckpointFormat):
name: typing.ClassVar[str] = "gpt_oss"
Loading