Upload quantize_config.json with huggingface_hub
Browse files- quantize_config.json +25 -0
quantize_config.json
ADDED
|
@@ -0,0 +1,25 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bits": 4,
|
| 3 |
+
"group_size": 128,
|
| 4 |
+
"sym": true,
|
| 5 |
+
"data_type": "fp8_gaudi3_to_int_sym",
|
| 6 |
+
"enable_quanted_input": true,
|
| 7 |
+
"enable_minmax_tuning": true,
|
| 8 |
+
"seqlen": 512,
|
| 9 |
+
"batch_size": 4,
|
| 10 |
+
"scale_dtype": "torch.bfloat16",
|
| 11 |
+
"lr": 0.005,
|
| 12 |
+
"minmax_lr": 0.005,
|
| 13 |
+
"gradient_accumulate_steps": 1,
|
| 14 |
+
"iters": 50,
|
| 15 |
+
"amp": true,
|
| 16 |
+
"nsamples": 512,
|
| 17 |
+
"low_gpu_mem_usage": true,
|
| 18 |
+
"to_quant_block_names": null,
|
| 19 |
+
"enable_norm_bias_tuning": false,
|
| 20 |
+
"autoround_version": "0.4.5",
|
| 21 |
+
"quant_method": "gptq",
|
| 22 |
+
"desc_act": false,
|
| 23 |
+
"true_sequential": false,
|
| 24 |
+
"damp_percent": 0.01
|
| 25 |
+
}
|