TharunSivamani commited on
Commit
c438ede
1 Parent(s): db2746d

config msg

Browse files
Files changed (1) hide show
  1. config.json +162 -0
config.json ADDED
@@ -0,0 +1,162 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "version": "0.13.0.dev2024090300",
3
+ "pretrained_config": {
4
+ "architecture": "BloomForCausalLM",
5
+ "dtype": "float16",
6
+ "vocab_size": 250880,
7
+ "hidden_size": 1024,
8
+ "num_hidden_layers": 24,
9
+ "num_attention_heads": 16,
10
+ "hidden_act": "gelu",
11
+ "logits_dtype": "float32",
12
+ "norm_epsilon": 1e-05,
13
+ "position_embedding_type": "alibi",
14
+ "max_position_embeddings": null,
15
+ "num_key_value_heads": 16,
16
+ "intermediate_size": 4096,
17
+ "mapping": {
18
+ "world_size": 1,
19
+ "gpus_per_node": 8,
20
+ "cp_size": 1,
21
+ "tp_size": 1,
22
+ "pp_size": 1,
23
+ "moe_tp_size": 1,
24
+ "moe_ep_size": 1
25
+ },
26
+ "quantization": {
27
+ "quant_algo": null,
28
+ "kv_cache_quant_algo": null,
29
+ "group_size": 128,
30
+ "smoothquant_val": 0.5,
31
+ "clamp_val": null,
32
+ "has_zero_point": false,
33
+ "pre_quant_scale": false,
34
+ "exclude_modules": null
35
+ },
36
+ "use_parallel_embedding": false,
37
+ "embedding_sharding_dim": 0,
38
+ "share_embedding_table": false,
39
+ "head_size": 64,
40
+ "qk_layernorm": false
41
+ },
42
+ "build_config": {
43
+ "max_input_len": 1024,
44
+ "max_seq_len": 2048,
45
+ "opt_batch_size": null,
46
+ "max_batch_size": 256,
47
+ "max_beam_width": 1,
48
+ "max_num_tokens": 8192,
49
+ "opt_num_tokens": 256,
50
+ "max_prompt_embedding_table_size": 0,
51
+ "kv_cache_type": "PAGED",
52
+ "gather_context_logits": false,
53
+ "gather_generation_logits": false,
54
+ "strongly_typed": true,
55
+ "builder_opt": null,
56
+ "force_num_profiles": null,
57
+ "profiling_verbosity": "layer_names_only",
58
+ "enable_debug_output": false,
59
+ "max_draft_len": 0,
60
+ "speculative_decoding_mode": 1,
61
+ "use_refit": false,
62
+ "input_timing_cache": null,
63
+ "output_timing_cache": "model.cache",
64
+ "lora_config": {
65
+ "lora_dir": [],
66
+ "lora_ckpt_source": "hf",
67
+ "max_lora_rank": 64,
68
+ "lora_target_modules": [],
69
+ "trtllm_modules_to_hf_modules": {}
70
+ },
71
+ "auto_parallel_config": {
72
+ "world_size": 1,
73
+ "gpus_per_node": 8,
74
+ "cluster_key": "NVIDIA-GeForce-RTX-3060-Laptop-GPU",
75
+ "cluster_info": {
76
+ "inter_node_bw_per_device": 25,
77
+ "intra_node_bw_per_device": 8,
78
+ "inter_node_latency": 10,
79
+ "intra_node_latency": 10,
80
+ "intra_node_sharp": false,
81
+ "inter_node_sharp": true,
82
+ "memory_bw": 336,
83
+ "memory_budget_per_device": 6,
84
+ "math_throughput": {
85
+ "int4": 258,
86
+ "int8": 129,
87
+ "fp8": 0,
88
+ "float16": 64,
89
+ "bfloat16": 64,
90
+ "float32": 32
91
+ },
92
+ "memory_efficiency": 1.0,
93
+ "math_efficiency": 1.0,
94
+ "communication_efficiency": 1.0
95
+ },
96
+ "sharding_cost_model": "alpha_beta",
97
+ "comm_cost_model": "alpha_beta",
98
+ "enable_pipeline_parallelism": false,
99
+ "enable_shard_unbalanced_shape": false,
100
+ "enable_shard_dynamic_shape": false,
101
+ "enable_reduce_scatter": true,
102
+ "builder_flags": null,
103
+ "debug_mode": false,
104
+ "infer_shape": true,
105
+ "validation_mode": false,
106
+ "same_buffer_io": {
107
+ "past_key_value_(\\d+)": "present_key_value_\\1"
108
+ },
109
+ "same_spec_io": {},
110
+ "sharded_io_allowlist": [
111
+ "past_key_value_\\d+",
112
+ "present_key_value_\\d*"
113
+ ],
114
+ "fill_weights": false,
115
+ "parallel_config_cache": null,
116
+ "profile_cache": null,
117
+ "dump_path": null,
118
+ "debug_outputs": []
119
+ },
120
+ "weight_sparsity": false,
121
+ "weight_streaming": false,
122
+ "plugin_config": {
123
+ "dtype": "float16",
124
+ "bert_attention_plugin": "auto",
125
+ "gpt_attention_plugin": "auto",
126
+ "gemm_plugin": "float16",
127
+ "gemm_swiglu_plugin": null,
128
+ "fp8_rowwise_gemm_plugin": null,
129
+ "smooth_quant_gemm_plugin": null,
130
+ "identity_plugin": null,
131
+ "layernorm_quantization_plugin": null,
132
+ "rmsnorm_quantization_plugin": null,
133
+ "nccl_plugin": null,
134
+ "lookup_plugin": null,
135
+ "lora_plugin": null,
136
+ "weight_only_groupwise_quant_matmul_plugin": null,
137
+ "weight_only_quant_matmul_plugin": null,
138
+ "quantize_per_token_plugin": false,
139
+ "quantize_tensor_plugin": false,
140
+ "moe_plugin": "auto",
141
+ "mamba_conv1d_plugin": "auto",
142
+ "low_latency_gemm_plugin": null,
143
+ "context_fmha": true,
144
+ "bert_context_fmha_fp32_acc": false,
145
+ "paged_kv_cache": true,
146
+ "remove_input_padding": true,
147
+ "reduce_fusion": false,
148
+ "enable_xqa": true,
149
+ "tokens_per_block": 64,
150
+ "use_paged_context_fmha": false,
151
+ "use_fp8_context_fmha": false,
152
+ "multiple_profiles": false,
153
+ "paged_state": false,
154
+ "streamingllm": false,
155
+ "manage_weights": false,
156
+ "use_fused_mlp": true
157
+ },
158
+ "use_strip_plan": false,
159
+ "max_encoder_input_len": 1024,
160
+ "use_fused_mlp": "enable"
161
+ }
162
+ }