Evaluate vLLM 0.20 profiles against Frontier
This commit is contained in:
@@ -0,0 +1,87 @@
|
||||
{
|
||||
"comparison_contract": {
|
||||
"calibration_a_tp": 1.0,
|
||||
"fixture": "fidelity_p1_tp1_mns64_low1",
|
||||
"scheduler_state": "batch_id=0, layer_id=0 before profile-dependent trajectories diverge"
|
||||
},
|
||||
"layer_component_sum_ms": {
|
||||
"historical_profile": 1.148334,
|
||||
"new_over_old": 2.700194368537377,
|
||||
"vllm020_profile": 3.1007249999999997
|
||||
},
|
||||
"rows": [
|
||||
{
|
||||
"component": "input_layernorm",
|
||||
"historical_profile_ms": 0.010619,
|
||||
"new_over_old": 1.955833882663151,
|
||||
"vllm020_profile_ms": 0.020769
|
||||
},
|
||||
{
|
||||
"component": "attn_pre_proj",
|
||||
"historical_profile_ms": 0.204389,
|
||||
"new_over_old": 1.0199276869107439,
|
||||
"vllm020_profile_ms": 0.208462
|
||||
},
|
||||
{
|
||||
"component": "attn_rope",
|
||||
"historical_profile_ms": 0.018908,
|
||||
"new_over_old": 7.394171779141103,
|
||||
"vllm020_profile_ms": 0.139809
|
||||
},
|
||||
{
|
||||
"component": "attn_kv_cache_save",
|
||||
"historical_profile_ms": 0.014226,
|
||||
"new_over_old": 1.427386475467454,
|
||||
"vllm020_profile_ms": 0.020306
|
||||
},
|
||||
{
|
||||
"component": "attn_prefill",
|
||||
"historical_profile_ms": 0.107759,
|
||||
"new_over_old": 2.670709639102071,
|
||||
"vllm020_profile_ms": 0.287793
|
||||
},
|
||||
{
|
||||
"component": "attn_decode",
|
||||
"historical_profile_ms": 0.0,
|
||||
"new_over_old": null,
|
||||
"vllm020_profile_ms": 0.0
|
||||
},
|
||||
{
|
||||
"component": "attn_post_proj",
|
||||
"historical_profile_ms": 0.134153,
|
||||
"new_over_old": 0.967559428413826,
|
||||
"vllm020_profile_ms": 0.129801
|
||||
},
|
||||
{
|
||||
"component": "post_attention_layernorm",
|
||||
"historical_profile_ms": 0.011481,
|
||||
"new_over_old": 0.9000087100426792,
|
||||
"vllm020_profile_ms": 0.010333
|
||||
},
|
||||
{
|
||||
"component": "moe_gating_linear",
|
||||
"historical_profile_ms": 0.011802,
|
||||
"new_over_old": 2.2642772411455687,
|
||||
"vllm020_profile_ms": 0.026723
|
||||
},
|
||||
{
|
||||
"component": "moe_gating_routing_topk",
|
||||
"historical_profile_ms": 0.009662,
|
||||
"new_over_old": 2.2922790312564687,
|
||||
"vllm020_profile_ms": 0.022148
|
||||
},
|
||||
{
|
||||
"component": "moe_shuffling",
|
||||
"historical_profile_ms": 0.030431,
|
||||
"new_over_old": 0.0,
|
||||
"vllm020_profile_ms": 0.0
|
||||
},
|
||||
{
|
||||
"component": "moe_grouped_gemm",
|
||||
"historical_profile_ms": 0.594904,
|
||||
"new_over_old": 3.756204362384519,
|
||||
"vllm020_profile_ms": 2.234581
|
||||
}
|
||||
],
|
||||
"schema": "frontier-initial-op-trace-delta.v1"
|
||||
}
|
||||
Reference in New Issue
Block a user