hizrianraz's picture
docs tip: DFlash DO_NOT_PROMOTE + last-green + jury + analyze prompts (4e6496a)
c431598 verified
Raw
History Blame Contribute Delete
1.02 kB
{
"ts": 1785314162.0283267,
"mode": "dflash",
"spec": {
"type": "draft-dflash",
"n_max": 7,
"block_size": 16
},
"base_url": "http://127.0.0.1:8000/v1",
"model": "local-laguna",
"rows": [
{
"label": "gen8_short",
"prompt_tokens": 48,
"completion_tokens": 3,
"latency_s": 0.483,
"tok_s": 6.212,
"content_head": "OK"
},
{
"label": "gen128",
"prompt_tokens": 58,
"completion_tokens": 128,
"latency_s": 8.374,
"tok_s": 15.286,
"content_head": "Speculative decoding is a model inference optimization technique that accelerates text generation by using a fast, light"
},
{
"label": "gen256",
"prompt_tokens": 61,
"completion_tokens": 256,
"latency_s": 14.762,
"tok_s": 17.342,
"content_head": "## Block-Diffusion vs EAGLE-3: Practical Serving Comparison\n\n### **Block-Diffusion Draft Models**\n\n**What it is:** Uses "
}
],
"schema": "laguna.dflash_throughput/v1"
}