Skip to content

update minicpm4_6 #138

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

Merged
Jintao-Huang merged 1 commit into from
Jun 28, 2026
Merged

update minicpm4_6 #138

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
3 changes: 2 additions & 1 deletion src/mcore_bridge/config/parser.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -152,7 +152,8 @@ def hf_to_mcore_config(hf_config: PretrainedConfig) -> Dict[str, Any]:
moe_n_hash_layers = res.pop('moe_n_hash_layers', None)
rope_scaling = res.get('rope_scaling') or {}
if llm_model_type in {'qwen3', 'qwen3_moe', 'qwen3_next'} or hf_model_type in {
'qwen3_omni_moe', 'qwen3_omni', 'qwen3_vl', 'qwen3_vl_moe', 'qwen3_5', 'qwen3_5_moe', 'llavaonevision1_5', 'minicpmv4_6'
'qwen3_omni_moe', 'qwen3_omni', 'qwen3_vl', 'qwen3_vl_moe', 'qwen3_5', 'qwen3_5_moe', 'llavaonevision1_5',
'minicpmv4_6'
}:
res['qk_layernorm'] = True
if llm_model_type in {'qwen2_moe', 'qwen3_moe', 'qwen3_next'
Expand Down
3 changes: 2 additions & 1 deletion src/mcore_bridge/model/mm_gpts/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,2 +1,3 @@
# Copyright (c) ModelScope Contributors. All rights reserved.
from . import gemma4, glm, internvl, kimi_vl, llama4, llava, minicpmv4_6, qwen, qwen3_5, qwen3_5_gdn, qwen3_asr, qwen3_omni, qwen3_vl
from . import (gemma4, glm, internvl, kimi_vl, llama4, llava, minicpmv4_6, qwen, qwen3_5, qwen3_5_gdn, qwen3_asr,
qwen3_omni, qwen3_vl)
33 changes: 14 additions & 19 deletions src/mcore_bridge/model/mm_gpts/minicpmv4_6.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,11 +17,8 @@ class MiniCPMV46Vit(HuggingFaceVit):
_aligner = ['merger']

def prepare_model(self, hf_config: PretrainedConfig):
from transformers.models.minicpmv4_6.modeling_minicpmv4_6 import (
MiniCPMV4_6VisionModel,
MiniCPMV4_6Merger,
MiniCPMV4_6Model
)
from transformers.models.minicpmv4_6.modeling_minicpmv4_6 import (MiniCPMV4_6Merger, MiniCPMV4_6Model,
MiniCPMV4_6VisionModel)
self.vision_tower = MiniCPMV4_6VisionModel._from_config(hf_config.vision_config)
self.merger = MiniCPMV4_6Merger(hf_config).to(dtype=self.vision_tower.dtype)
self.model_cls = MiniCPMV4_6Model
Expand All @@ -37,10 +34,8 @@ def get_inputs_embeds(self, inputs_embeds, **kwargs):
if pixel_values is None and pixel_values_videos is None:
patch_size = hf_config.vision_config.patch_size
dummy_pv = torch.zeros(
1, 3, 4 * patch_size, 4 * patch_size,
device=inputs_embeds.device, dtype=self.vision_tower.dtype)
dummy_ts = torch.tensor(
[[4, 4]], device=inputs_embeds.device, dtype=torch.int32)
1, 3, 4 * patch_size, 4 * patch_size, device=inputs_embeds.device, dtype=self.vision_tower.dtype)
dummy_ts = torch.tensor([[4, 4]], device=inputs_embeds.device, dtype=torch.int32)
with self.patch_hf_config():
vision_output = self.model_cls.get_image_features(self, dummy_pv, dummy_ts)
image_embeds = torch.cat(vision_output.pooler_output, dim=0)
Expand All @@ -52,11 +47,11 @@ def get_inputs_embeds(self, inputs_embeds, **kwargs):
vision_output = self.model_cls.get_image_features(
self, pixel_values[:1].to(dtype=self.vision_tower.dtype), target_sizes)
image_features = (
torch.cat(vision_output.pooler_output, dim=0)
.to(device=inputs_embeds.device, dtype=inputs_embeds.dtype)
.repeat(num_beams, 1))
mask = self.model_cls.get_placeholder_mask(
self, input_ids, inputs_embeds, image_features, hf_config.image_token_id)
torch.cat(vision_output.pooler_output,
dim=0).to(device=inputs_embeds.device,
dtype=inputs_embeds.dtype).repeat(num_beams, 1))
mask = self.model_cls.get_placeholder_mask(self, input_ids, inputs_embeds, image_features,
hf_config.image_token_id)
inputs_embeds = inputs_embeds.masked_scatter(mask, image_features)

if pixel_values_videos is not None:
Expand All @@ -65,11 +60,11 @@ def get_inputs_embeds(self, inputs_embeds, **kwargs):
vision_output = self.model_cls.get_video_features(
self, pixel_values_videos[:1].to(dtype=self.vision_tower.dtype), target_sizes_videos)
video_features = (
torch.cat(vision_output.pooler_output, dim=0)
.to(device=inputs_embeds.device, dtype=inputs_embeds.dtype)
.repeat(num_beams, 1))
mask = self.model_cls.get_placeholder_mask(
self, input_ids, inputs_embeds, video_features, hf_config.video_token_id)
torch.cat(vision_output.pooler_output,
dim=0).to(device=inputs_embeds.device,
dtype=inputs_embeds.dtype).repeat(num_beams, 1))
mask = self.model_cls.get_placeholder_mask(self, input_ids, inputs_embeds, video_features,
hf_config.video_token_id)
inputs_embeds = inputs_embeds.masked_scatter(mask, video_features)

return inputs_embeds
Expand Down
Loading