huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-09-10 20:04:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

bump torch to more recent version

Browse Source
This commit is contained in:
Felix Marty 2024-06-28 13:10:43 +00:00
parent 87db820627
commit 3d50ff71b7
4 changed files with 11 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
.github/workflows/build.yaml vendored
View File

@ -251,7 +251,7 @@ jobs:
- name: Run tests
run: |
export DOCKER_DEVICES=${{ needs.build-and-push.outputs.docker_devices }}
export HUGGING_FACE_HUB_TOKEN=${{ secrets.HF_TOKEN }}
export HF_TOKEN=${{ secrets.HF_TOKEN }}
export DOCKER_IMAGE=${{ needs.build-and-push.outputs.docker_image }}
echo "DOCKER_IMAGE:"

5
Dockerfile_amd
View File

@ -96,7 +96,10 @@ RUN pip uninstall -y triton && \
cd triton/python && \
pip install .
RUN git clone --depth 1 --recursive --single-branch --branch 2.3-patched https://github.com/fxmarty/pytorch.git pytorch && cd pytorch && pip install -r requirements.txt --no-cache-dir
RUN git clone --depth 1 --recursive --single-branch --branch 2.3-patched https://github.com/fxmarty/pytorch.git pytorch && \
cd pytorch && \
git checkout adbb514d024d4ba1b6f9c126d321fd5e7597e966 && \
pip install -r requirements.txt --no-cache-dir
ARG _GLIBCXX_USE_CXX11_ABI="1"
ARG CMAKE_PREFIX_PATH="/opt/conda"

4
server/text_generation_server/models/custom_modeling/flash_llama_modeling.py
View File

@ -309,7 +309,9 @@ class LlamaMLP(nn.Module):
dtype=hidden_states.dtype,
device="cuda",
)
_custom_C.LLMM_Silu(self.gate_up_proj.linear.weight, hidden_states, out, 8)
_custom_C.LLMM_Silu(
self.gate_up_proj.base_layer.linear.weight, hidden_states, out, 8
)
return self.down_proj(out, adapter_data)
else:
gate_up_states = self.gate_up_proj(hidden_states, adapter_data)

4
server/text_generation_server/models/custom_modeling/flash_mistral_modeling.py
View File

@ -314,7 +314,9 @@ class MistralMLP(nn.Module):
dtype=hidden_states.dtype,
device="cuda",
)
_custom_C.LLMM_Silu(self.gate_up_proj.linear.weight, hidden_states, out, 8)
_custom_C.LLMM_Silu(
self.gate_up_proj.base_layer.linear.weight, hidden_states, out, 8
)
return self.down_proj(out, adapter_data)
else:
gate_up_states = self.gate_up_proj(hidden_states, adapter_data)