Commit
•
2a84df9
1
Parent(s):
82ba527
Upload folder using huggingface_hub
Browse files- config.json +148 -0
config.json
ADDED
@@ -0,0 +1,148 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"version": "0.11.0.dev2024061800",
|
3 |
+
"pretrained_config": {
|
4 |
+
"mlp_bias": false,
|
5 |
+
"attn_bias": false,
|
6 |
+
"rotary_base": 1000000.0,
|
7 |
+
"rotary_scaling": null,
|
8 |
+
"residual_mlp": false,
|
9 |
+
"disable_weight_only_quant_plugin": false,
|
10 |
+
"moe": {
|
11 |
+
"num_experts": 0,
|
12 |
+
"top_k": 0,
|
13 |
+
"normalization_mode": null
|
14 |
+
},
|
15 |
+
"architecture": "LlamaForCausalLM",
|
16 |
+
"dtype": "float16",
|
17 |
+
"vocab_size": 32000,
|
18 |
+
"hidden_size": 4096,
|
19 |
+
"num_hidden_layers": 32,
|
20 |
+
"num_attention_heads": 32,
|
21 |
+
"hidden_act": "silu",
|
22 |
+
"logits_dtype": "float32",
|
23 |
+
"norm_epsilon": 1e-05,
|
24 |
+
"position_embedding_type": "rope_gpt_neox",
|
25 |
+
"max_position_embeddings": 32768,
|
26 |
+
"num_key_value_heads": 8,
|
27 |
+
"intermediate_size": 14336,
|
28 |
+
"mapping": {
|
29 |
+
"world_size": 2,
|
30 |
+
"gpus_per_node": 8,
|
31 |
+
"tp_size": 2,
|
32 |
+
"pp_size": 1,
|
33 |
+
"moe_tp_size": 2,
|
34 |
+
"moe_ep_size": 1
|
35 |
+
},
|
36 |
+
"quantization": {
|
37 |
+
"quant_algo": null,
|
38 |
+
"kv_cache_quant_algo": null,
|
39 |
+
"group_size": 128,
|
40 |
+
"smoothquant_val": null,
|
41 |
+
"has_zero_point": false,
|
42 |
+
"pre_quant_scale": false,
|
43 |
+
"exclude_modules": null
|
44 |
+
},
|
45 |
+
"use_parallel_embedding": false,
|
46 |
+
"embedding_sharding_dim": 0,
|
47 |
+
"share_embedding_table": false,
|
48 |
+
"head_size": 128,
|
49 |
+
"qk_layernorm": false
|
50 |
+
},
|
51 |
+
"build_config": {
|
52 |
+
"max_input_len": 1024,
|
53 |
+
"max_seq_len": 2048,
|
54 |
+
"opt_batch_size": null,
|
55 |
+
"max_batch_size": 64,
|
56 |
+
"max_beam_width": 1,
|
57 |
+
"max_num_tokens": 8192,
|
58 |
+
"opt_num_tokens": 64,
|
59 |
+
"max_prompt_embedding_table_size": 0,
|
60 |
+
"gather_context_logits": false,
|
61 |
+
"gather_generation_logits": false,
|
62 |
+
"strongly_typed": true,
|
63 |
+
"builder_opt": null,
|
64 |
+
"profiling_verbosity": "layer_names_only",
|
65 |
+
"enable_debug_output": false,
|
66 |
+
"max_draft_len": 0,
|
67 |
+
"speculative_decoding_mode": 1,
|
68 |
+
"use_refit": false,
|
69 |
+
"input_timing_cache": null,
|
70 |
+
"output_timing_cache": "model.cache",
|
71 |
+
"lora_config": {
|
72 |
+
"lora_dir": [],
|
73 |
+
"lora_ckpt_source": "hf",
|
74 |
+
"max_lora_rank": 64,
|
75 |
+
"lora_target_modules": [],
|
76 |
+
"trtllm_modules_to_hf_modules": {}
|
77 |
+
},
|
78 |
+
"auto_parallel_config": {
|
79 |
+
"world_size": 1,
|
80 |
+
"gpus_per_node": 8,
|
81 |
+
"cluster_key": "A100-SXM-80GB",
|
82 |
+
"cluster_info": null,
|
83 |
+
"sharding_cost_model": "alpha_beta",
|
84 |
+
"comm_cost_model": "alpha_beta",
|
85 |
+
"enable_pipeline_parallelism": false,
|
86 |
+
"enable_shard_unbalanced_shape": false,
|
87 |
+
"enable_shard_dynamic_shape": false,
|
88 |
+
"enable_reduce_scatter": true,
|
89 |
+
"builder_flags": null,
|
90 |
+
"debug_mode": false,
|
91 |
+
"infer_shape": true,
|
92 |
+
"validation_mode": false,
|
93 |
+
"same_buffer_io": {
|
94 |
+
"past_key_value_(\\d+)": "present_key_value_\\1"
|
95 |
+
},
|
96 |
+
"same_spec_io": {},
|
97 |
+
"sharded_io_allowlist": [
|
98 |
+
"past_key_value_\\d+",
|
99 |
+
"present_key_value_\\d*"
|
100 |
+
],
|
101 |
+
"fill_weights": false,
|
102 |
+
"parallel_config_cache": null,
|
103 |
+
"profile_cache": null,
|
104 |
+
"dump_path": null,
|
105 |
+
"debug_outputs": []
|
106 |
+
},
|
107 |
+
"weight_sparsity": false,
|
108 |
+
"weight_streaming": false,
|
109 |
+
"plugin_config": {
|
110 |
+
"dtype": "float16",
|
111 |
+
"bert_attention_plugin": "auto",
|
112 |
+
"gpt_attention_plugin": "auto",
|
113 |
+
"gemm_plugin": "float16",
|
114 |
+
"gemm_swiglu_plugin": null,
|
115 |
+
"smooth_quant_gemm_plugin": null,
|
116 |
+
"identity_plugin": null,
|
117 |
+
"layernorm_quantization_plugin": null,
|
118 |
+
"rmsnorm_quantization_plugin": null,
|
119 |
+
"nccl_plugin": "float16",
|
120 |
+
"lookup_plugin": null,
|
121 |
+
"lora_plugin": null,
|
122 |
+
"weight_only_groupwise_quant_matmul_plugin": null,
|
123 |
+
"weight_only_quant_matmul_plugin": null,
|
124 |
+
"quantize_per_token_plugin": false,
|
125 |
+
"quantize_tensor_plugin": false,
|
126 |
+
"moe_plugin": "auto",
|
127 |
+
"mamba_conv1d_plugin": "auto",
|
128 |
+
"context_fmha": true,
|
129 |
+
"context_fmha_fp32_acc": false,
|
130 |
+
"paged_kv_cache": true,
|
131 |
+
"remove_input_padding": true,
|
132 |
+
"use_custom_all_reduce": true,
|
133 |
+
"reduce_fusion": false,
|
134 |
+
"multi_block_mode": false,
|
135 |
+
"enable_xqa": true,
|
136 |
+
"attention_qk_half_accumulation": false,
|
137 |
+
"tokens_per_block": 64,
|
138 |
+
"use_paged_context_fmha": true,
|
139 |
+
"use_fp8_context_fmha": false,
|
140 |
+
"multiple_profiles": false,
|
141 |
+
"paged_state": true,
|
142 |
+
"streamingllm": false
|
143 |
+
},
|
144 |
+
"use_strip_plan": false,
|
145 |
+
"max_encoder_input_len": 1024,
|
146 |
+
"use_fused_mlp": false
|
147 |
+
}
|
148 |
+
}
|