Hugging Face's logo

pandyamarut
/
llama-3-8B-TRT-LLM-H100-SXM-TP1

Model card Files Community
pandyamarut commited on
Commit
cdccf93
·
verified ·
1 Parent(s): b714cd4

Upload config.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. config.json +162 -0
config.json ADDED
@@ -0,0 +1,162 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "version": "0.15.0.dev2024101500",
3
+ "pretrained_config": {
4
+ "mlp_bias": false,
5
+ "attn_bias": false,
6
+ "rotary_base": 500000.0,
7
+ "rotary_scaling": null,
8
+ "residual_mlp": false,
9
+ "disable_weight_only_quant_plugin": false,
10
+ "moe": {
11
+ "num_experts": 0,
12
+ "moe_intermediate_size": 0,
13
+ "num_shared_experts": 0,
14
+ "top_k": 0,
15
+ "normalization_mode": null,
16
+ "sparse_mixer_epsilon": 0.01,
17
+ "tp_mode": 0
18
+ },
19
+ "remove_duplicated_kv_heads": false,
20
+ "fc_after_embed": false,
21
+ "use_input_layernorm_in_first_layer": true,
22
+ "architecture": "LlamaForCausalLM",
23
+ "dtype": "bfloat16",
24
+ "vocab_size": 128256,
25
+ "hidden_size": 4096,
26
+ "num_hidden_layers": 32,
27
+ "num_attention_heads": 32,
28
+ "hidden_act": "silu",
29
+ "logits_dtype": "float32",
30
+ "norm_epsilon": 1e-05,
31
+ "position_embedding_type": "rope_gpt_neox",
32
+ "max_position_embeddings": 8192,
33
+ "num_key_value_heads": 8,
34
+ "intermediate_size": 14336,
35
+ "mapping": {
36
+ "world_size": 1,
37
+ "gpus_per_node": 8,
38
+ "cp_size": 1,
39
+ "tp_size": 1,
40
+ "pp_size": 1,
41
+ "moe_tp_size": 1,
42
+ "moe_ep_size": 1
43
+ },
44
+ "quantization": {
45
+ "quant_algo": null,
46
+ "kv_cache_quant_algo": null,
47
+ "group_size": 128,
48
+ "smoothquant_val": 0.5,
49
+ "clamp_val": null,
50
+ "has_zero_point": false,
51
+ "pre_quant_scale": false,
52
+ "exclude_modules": null
53
+ },
54
+ "use_parallel_embedding": true,
55
+ "embedding_sharding_dim": 0,
56
+ "share_embedding_table": false,
57
+ "head_size": 128,
58
+ "qk_layernorm": false
59
+ },
60
+ "build_config": {
61
+ "max_input_len": 256,
62
+ "max_seq_len": 512,
63
+ "opt_batch_size": 8,
64
+ "max_batch_size": 8,
65
+ "max_beam_width": 1,
66
+ "max_num_tokens": 4096,
67
+ "opt_num_tokens": 8,
68
+ "max_prompt_embedding_table_size": 0,
69
+ "kv_cache_type": "PAGED",
70
+ "gather_context_logits": false,
71
+ "gather_generation_logits": false,
72
+ "strongly_typed": true,
73
+ "force_num_profiles": null,
74
+ "profiling_verbosity": "layer_names_only",
75
+ "enable_debug_output": false,
76
+ "max_draft_len": 0,
77
+ "speculative_decoding_mode": 1,
78
+ "use_refit": false,
79
+ "input_timing_cache": null,
80
+ "output_timing_cache": null,
81
+ "lora_config": {
82
+ "lora_dir": [],
83
+ "lora_ckpt_source": "hf",
84
+ "max_lora_rank": 64,
85
+ "lora_target_modules": [],
86
+ "trtllm_modules_to_hf_modules": {}
87
+ },
88
+ "auto_parallel_config": {
89
+ "world_size": 1,
90
+ "gpus_per_node": 8,
91
+ "cluster_key": "H100-PCIe",
92
+ "cluster_info": null,
93
+ "sharding_cost_model": "alpha_beta",
94
+ "comm_cost_model": "alpha_beta",
95
+ "enable_pipeline_parallelism": false,
96
+ "enable_shard_unbalanced_shape": false,
97
+ "enable_shard_dynamic_shape": false,
98
+ "enable_reduce_scatter": true,
99
+ "builder_flags": null,
100
+ "debug_mode": false,
101
+ "infer_shape": true,
102
+ "validation_mode": false,
103
+ "same_buffer_io": {
104
+ "past_key_value_(\\d+)": "present_key_value_\\1"
105
+ },
106
+ "same_spec_io": {},
107
+ "sharded_io_allowlist": [
108
+ "past_key_value_\\d+",
109
+ "present_key_value_\\d*"
110
+ ],
111
+ "fill_weights": false,
112
+ "parallel_config_cache": null,
113
+ "profile_cache": null,
114
+ "dump_path": null,
115
+ "debug_outputs": []
116
+ },
117
+ "weight_sparsity": false,
118
+ "weight_streaming": false,
119
+ "plugin_config": {
120
+ "dtype": "bfloat16",
121
+ "bert_attention_plugin": "auto",
122
+ "gpt_attention_plugin": "auto",
123
+ "gemm_plugin": null,
124
+ "gemm_swiglu_plugin": null,
125
+ "fp8_rowwise_gemm_plugin": null,
126
+ "smooth_quant_gemm_plugin": null,
127
+ "identity_plugin": null,
128
+ "layernorm_quantization_plugin": null,
129
+ "rmsnorm_quantization_plugin": null,
130
+ "nccl_plugin": null,
131
+ "lookup_plugin": null,
132
+ "lora_plugin": null,
133
+ "weight_only_groupwise_quant_matmul_plugin": null,
134
+ "weight_only_quant_matmul_plugin": null,
135
+ "smooth_quant_plugins": true,
136
+ "quantize_per_token_plugin": false,
137
+ "quantize_tensor_plugin": false,
138
+ "moe_plugin": "auto",
139
+ "mamba_conv1d_plugin": "auto",
140
+ "low_latency_gemm_plugin": null,
141
+ "context_fmha": true,
142
+ "bert_context_fmha_fp32_acc": false,
143
+ "paged_kv_cache": true,
144
+ "remove_input_padding": true,
145
+ "reduce_fusion": false,
146
+ "enable_xqa": true,
147
+ "tokens_per_block": 64,
148
+ "use_paged_context_fmha": false,
149
+ "use_fp8_context_fmha": false,
150
+ "multiple_profiles": false,
151
+ "paged_state": false,
152
+ "streamingllm": false,
153
+ "manage_weights": false,
154
+ "use_fused_mlp": true,
155
+ "pp_reduce_scatter": false
156
+ },
157
+ "use_strip_plan": false,
158
+ "max_encoder_input_len": 1,
159
+ "use_fused_mlp": false,
160
+ "monitor_memory": false
161
+ }
162
+ }