Upload ms-swift/examples/export/quantize/moe/gptq.sh with huggingface_hub
Browse files
ms-swift/examples/export/quantize/moe/gptq.sh
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# 2 * 80GB
|
| 2 |
+
OMP_NUM_THREADS=14 \
|
| 3 |
+
CUDA_VISIBLE_DEVICES=0,1 \
|
| 4 |
+
swift export \
|
| 5 |
+
--model Qwen/Qwen2-57B-A14B-Instruct \
|
| 6 |
+
--dataset 'AI-ModelScope/alpaca-gpt4-data-zh#1000' \
|
| 7 |
+
'AI-ModelScope/alpaca-gpt4-data-en#1000' \
|
| 8 |
+
--quant_n_samples 512 \
|
| 9 |
+
--quant_batch_size 1 \
|
| 10 |
+
--max_length 4096 \
|
| 11 |
+
--quant_method gptq \
|
| 12 |
+
--quant_bits 4 \
|
| 13 |
+
--output_dir Qwen2-57B-A14B-Instruct-GPTQ-Int4
|