a0dd77bbd3ad5a61ab8695d30118db0d18729f34fdfbea8fb6d4d79e58647704
Browse files- config.json +6 -1
- smash_config.json +1 -1
config.json
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
{
|
2 |
"_attn_implementation_autoset": true,
|
3 |
-
"_name_or_path": "/tmp/models/
|
4 |
"architectures": [
|
5 |
"LlamaForCausalLM"
|
6 |
],
|
@@ -47,9 +47,14 @@
|
|
47 |
"tie_word_embeddings": true,
|
48 |
"torch_dtype": "bfloat16",
|
49 |
"transformers.js_config": {
|
|
|
50 |
"kv_cache_dtype": {
|
51 |
"fp16": "float16",
|
52 |
"q4f16": "float16"
|
|
|
|
|
|
|
|
|
53 |
}
|
54 |
},
|
55 |
"transformers_version": "4.48.2",
|
|
|
1 |
{
|
2 |
"_attn_implementation_autoset": true,
|
3 |
+
"_name_or_path": "/tmp/models/tmp3fjrifao/tmpizgz2wn6",
|
4 |
"architectures": [
|
5 |
"LlamaForCausalLM"
|
6 |
],
|
|
|
47 |
"tie_word_embeddings": true,
|
48 |
"torch_dtype": "bfloat16",
|
49 |
"transformers.js_config": {
|
50 |
+
"dtype": "q4",
|
51 |
"kv_cache_dtype": {
|
52 |
"fp16": "float16",
|
53 |
"q4f16": "float16"
|
54 |
+
},
|
55 |
+
"use_external_data_format": {
|
56 |
+
"model.onnx": true,
|
57 |
+
"model_fp16.onnx": true
|
58 |
}
|
59 |
},
|
60 |
"transformers_version": "4.48.2",
|
smash_config.json
CHANGED
@@ -11,7 +11,7 @@
|
|
11 |
"quant_hqq_weight_bits": 8,
|
12 |
"max_batch_size": 1,
|
13 |
"device": "cuda",
|
14 |
-
"cache_dir": "/tmp/models/
|
15 |
"task": "",
|
16 |
"save_load_fn": "hqq",
|
17 |
"save_load_fn_args": {},
|
|
|
11 |
"quant_hqq_weight_bits": 8,
|
12 |
"max_batch_size": 1,
|
13 |
"device": "cuda",
|
14 |
+
"cache_dir": "/tmp/models/tmp3fjrifao",
|
15 |
"task": "",
|
16 |
"save_load_fn": "hqq",
|
17 |
"save_load_fn_args": {},
|