Commit
•
4b63086
1
Parent(s):
3098727
sq version
Browse files
7b-sq-int8kv-tp1/config.json
ADDED
|
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"builder_config": {
|
| 3 |
+
"fp8": false,
|
| 4 |
+
"hidden_act": "silu",
|
| 5 |
+
"hidden_size": 4096,
|
| 6 |
+
"int8": true,
|
| 7 |
+
"max_batch_size": 8,
|
| 8 |
+
"max_input_len": 2048,
|
| 9 |
+
"max_num_tokens": null,
|
| 10 |
+
"max_output_len": 512,
|
| 11 |
+
"max_position_embeddings": 4096,
|
| 12 |
+
"name": "llama",
|
| 13 |
+
"num_heads": 32,
|
| 14 |
+
"num_kv_heads": 32,
|
| 15 |
+
"num_layers": 32,
|
| 16 |
+
"parallel_build": false,
|
| 17 |
+
"pipeline_parallel": 1,
|
| 18 |
+
"precision": "float16",
|
| 19 |
+
"quant_mode": 70,
|
| 20 |
+
"tensor_parallel": 1,
|
| 21 |
+
"use_refit": false,
|
| 22 |
+
"vocab_size": 32000
|
| 23 |
+
},
|
| 24 |
+
"plugin_config": {
|
| 25 |
+
"attention_qk_half_accumulation": false,
|
| 26 |
+
"bert_attention_plugin": false,
|
| 27 |
+
"context_fmha_type": 0,
|
| 28 |
+
"gemm_plugin": "float16",
|
| 29 |
+
"gpt_attention_plugin": "float16",
|
| 30 |
+
"identity_plugin": false,
|
| 31 |
+
"layernorm_plugin": false,
|
| 32 |
+
"layernorm_quantization_plugin": false,
|
| 33 |
+
"lookup_plugin": false,
|
| 34 |
+
"nccl_plugin": false,
|
| 35 |
+
"paged_kv_cache": false,
|
| 36 |
+
"quantize_per_token_plugin": true,
|
| 37 |
+
"quantize_tensor_plugin": true,
|
| 38 |
+
"remove_input_padding": false,
|
| 39 |
+
"rmsnorm_plugin": false,
|
| 40 |
+
"rmsnorm_quantization_plugin": "float16",
|
| 41 |
+
"smooth_quant_gemm_plugin": "float16",
|
| 42 |
+
"tokens_per_block": 0,
|
| 43 |
+
"use_custom_all_reduce": false,
|
| 44 |
+
"weight_only_groupwise_quant_matmul_plugin": false,
|
| 45 |
+
"weight_only_quant_matmul_plugin": false
|
| 46 |
+
}
|
| 47 |
+
}
|
7b-sq-int8kv-tp1/llama_float16_tp1_rank0.engine
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d3bd67a974e1ba09ce35f4de7d03010dfdb4089e72aec6a74c1c2e3714c0bca1
|
| 3 |
+
size 7006262500
|
7b-sq-int8kv-tp1/model.cache
ADDED
|
Binary file (114 kB). View file
|
|
|