Skip to content

Commit 2223d4b

Browse files
committed
Merge branch 'chenjiel/dsv32' of github.com:NVIDIA/TensorRT-Model-Optimizer into chenjiel/dsv32
2 parents ea9190e + 9b64663 commit 2223d4b

File tree

2 files changed

+18
-3
lines changed

2 files changed

+18
-3
lines changed

examples/deepseek/README.md

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -16,7 +16,7 @@ export FP4_QUANT_PATH={path_to_save_quantization_results}
1616
export HF_FP4_PATH={path_to_save_the_final_FP4_checkpoint}
1717
```
1818

19-
### DeepSeek V3, R1 and V3.1
19+
### DeepSeek V3 R1 V3.1
2020

2121
```bash
2222
# download the FP8 checkpoint from Hugginface. This is an example of DeepSeek-R1
@@ -37,7 +37,7 @@ git clone https://github.com/deepseek-ai/DeepSeek-V3.2-Exp.git && cd DeepSeek-V3
3737

3838
# Install requirements
3939
pip install git+https://github.com/Dao-AILab/fast-hadamard-transform.git
40-
pip install -r DeepSeek-V3.2-Exp/inference/requirements.txt
40+
pip install -r inference/requirements.txt
4141
```
4242

4343
### Convert the Checkpoint
@@ -51,7 +51,7 @@ python inference/convert.py --hf-ckpt-path $HF_FP8_CKPT --save-path $DS_CKPT --n
5151

5252
### Run the calibration scripts
5353

54-
DeepSeek V3
54+
DeepSeek V3, R1, V3.1
5555

5656
```bash
5757
torchrun --nproc-per-node 8 --master_port=12346 ptq.py --model_path $DS_CKPT --config DeepSeek-V3/inference/configs/config_671B.json --quant_cfg NVFP4_DEFAULT_CFG --output_path $FP4_QUANT_PATH

examples/deepseek/ds_kernel.py

Lines changed: 15 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -20,6 +20,21 @@
2020
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
2121
# SOFTWARE.
2222

23+
# SPDX-FileCopyrightText: Copyright (c) 2023-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
24+
# SPDX-License-Identifier: Apache-2.0
25+
#
26+
# Licensed under the Apache License, Version 2.0 (the "License");
27+
# you may not use this file except in compliance with the License.
28+
# You may obtain a copy of the License at
29+
#
30+
# http://www.apache.org/licenses/LICENSE-2.0
31+
#
32+
# Unless required by applicable law or agreed to in writing, software
33+
# distributed under the License is distributed on an "AS IS" BASIS,
34+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
35+
# See the License for the specific language governing permissions and
36+
# limitations under the License.
37+
2338
import torch
2439
import triton
2540
import triton.language as tl

0 commit comments

Comments
 (0)