10000 Add Qwen3 by yuanwu2017 · Pull Request #3229 · huggingface/text-generation-inference · GitHub
[go: up one dir, main page]
Skip to content

Add Qwen3 #3229

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
merged 9 commits into from
May 23, 2025
+386 −2
Merged

Add Qwen3 #3229

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
d704b0c
Add Qwen3
yuanwu2017 May 13, 2025
2a01478
Remove debug log
yuanwu2017 May 20, 2025
3d20c79
Merge branch 'huggingface:main' into qwen3
yuanwu2017 May 22, 2025
ad41abd
Add mark_step into qwen3
yuanwu2017 May 22, 2025
2e8d3e9
Add mark_step into llama4
yuanwu2017 May 22, 2025
346b6f7
Use the latest transformers
yuanwu2017 May 22, 2025
1ccf86c
Use the 4.52.2 transformers
yuanwu2017 May 22, 2025
5e1d1bf
Cannot use the latest transformers
yuanwu2017 May 22, 2025
cc3f612
Remove debug modification
yuanwu2017 May 22, 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
21 changes: 21 additions & 0 deletions backends/gaudi/server/text_generation_server/models/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -109,6 +109,9 @@
from text_generation_server.models.custom_modeling.flash_qwen2_modeling import (
Qwen2ForCausalLM,
)
from text_generation_server.models.custom_modeling.flash_qwen3_modeling import (
Qwen3ForCausalLM,
)
from text_generation_server.models.custom_modeling.flash_mistral_modeling import (
FlashMistralForCausalLM,
)
Expand Down Expand Up @@ -293,6 +296,12 @@ class ModelType(enum.Enum):
"name": "Qwen 2.5 VL",
"url": "https://huggingface.co/collections/Qwen/qwen25-66e81a666513e518adb90d9e",
}
QWEN3 = {
"type": "qwen3",
"name": "Qwen 3",
"url": "https://huggingface.co/collections/Qwen/qwen3-67dd247413f0e2e4f653967f",
}

GALACTICA = {
"type": "galactica",
"name": "Galactica",
Expand Down Expand Up @@ -791,6 +800,18 @@ def get_model(
config_class=Qwen2_5_VLConfig,
processor_class=Qwen2_5_VLProcessor,
)
elif model_type == QWEN3:
return FlashCausalLM(
model_id=model_id,
model_class=Qwen3ForCausalLM,
revision=revision,
quantize=quantize,
speculator=speculator,
dtype=dtype,
kv_cache_dtype=kv_cache_dtype,
trust_remote_code=trust_remote_code,
lora_adapter_ids=lora_adapter_ids,
)
elif model_type == MLLAMA:
return FlashMllamaCausalLM(
model_id=model_id,
Expand Down
6 changes: 6 additions & 0 deletions backends/gaudi/server/text_generation_server/models/custom_modeling/flash_llama4_modeling.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,7 @@
from torch import nn
import torch.nn.functional as F

import habana_frameworks.torch as htorch
from transformers.cache_utils import Cache
from transformers.activations import ACT2FN
from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS
Expand Down Expand Up @@ -567,6 +568,9 @@ def forward(
)

freqs_ci = self.rotary_emb(hidden_states, position_ids.view(bs, -1))
lazy_mode = htorch.utils.internal.is_lazy()
if lazy_mode:
htorch.core.mark_step()

for i, layer in enumerate(self.layers):
hidden_states = layer(
Expand All @@ -582,6 +586,8 @@ def forward(
position_ids=position_ids,
hpu_attention_meta=hpu_attention_meta,
)
if lazy_mode:
htorch.core.mark_step()

hidden_states, _ = self.norm(hidden_states)

Expand Down
Loading
0