204 lines
No EOL
5.6 KiB
JSON
204 lines
No EOL
5.6 KiB
JSON
{
|
|
"live_coder_8081": {
|
|
"port": 8081,
|
|
"n_ctx": 65536,
|
|
"total_slots": 1,
|
|
"tokenize_sample_tokens": 10,
|
|
"total_vram_mib": 26750,
|
|
"generation_speed_tps": 107.28,
|
|
"prompt_speed_tps": 156.8,
|
|
"timings": {
|
|
"cache_n": 13,
|
|
"prompt_n": 6,
|
|
"prompt_ms": 38.265,
|
|
"prompt_per_token_ms": 6.3775,
|
|
"prompt_per_second": 156.80125441003528,
|
|
"predicted_n": 128,
|
|
"predicted_ms": 1183.873,
|
|
"predicted_per_token_ms": 9.321834645669291,
|
|
"predicted_per_second": 107.27502020909337
|
|
}
|
|
},
|
|
"compressor_evaluation": {
|
|
"Qwen3-4B-2507_GPU": {
|
|
"name": "Qwen3-4B-2507",
|
|
"device": "GPU",
|
|
"ngl": 99,
|
|
"cold_start_sec": 3.01,
|
|
"process_vram_mib": 7446,
|
|
"avg_generation_tps": 137.18,
|
|
"avg_prompt_tps": 1595.19,
|
|
"evaluations": [
|
|
{
|
|
"prompt_id": "C01_code_repo_summary",
|
|
"content_preview": "Hermes Hub Router runs a FastAPI server on port 8765, routing requests to Ollama (port 11434), local llama.cpp (coder on port 8081, compress",
|
|
"gen_tps": 135.41,
|
|
"prompt_tps": 273.19
|
|
},
|
|
{
|
|
"prompt_id": "C02_security_audit_log",
|
|
"content_preview": "- **IP Address**: 192.168.1.105 - **Token Hash**: sha256:e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 - **Action**: ",
|
|
"gen_tps": 138.95,
|
|
"prompt_tps": 2917.19
|
|
}
|
|
],
|
|
"status": "OK"
|
|
},
|
|
"Qwen3-4B-2507_CPU": {
|
|
"name": "Qwen3-4B-2507",
|
|
"device": "CPU",
|
|
"ngl": 0,
|
|
"cold_start_sec": 12.01,
|
|
"process_vram_mib": 2648,
|
|
"avg_generation_tps": 8.42,
|
|
"avg_prompt_tps": 71.13,
|
|
"evaluations": [
|
|
{
|
|
"prompt_id": "C01_code_repo_summary",
|
|
"content_preview": "Hermes Hub Router runs a FastAPI server on port 8765, routing requests to Ollama (port 11434), local llama.cpp (coder on port 8081, compress",
|
|
"gen_tps": 8.47,
|
|
"prompt_tps": 42.47
|
|
},
|
|
{
|
|
"prompt_id": "C02_security_audit_log",
|
|
"content_preview": "- **IP Address**: 192.168.1.105 - **Token Hash**: sha256:e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 - **Action**: ",
|
|
"gen_tps": 8.36,
|
|
"prompt_tps": 99.78
|
|
}
|
|
],
|
|
"status": "OK"
|
|
},
|
|
"LFM2.5-2.6B_GPU": {
|
|
"name": "LFM2.5-2.6B",
|
|
"device": "GPU",
|
|
"ngl": 99,
|
|
"cold_start_sec": 6.01,
|
|
"process_vram_mib": 2562,
|
|
"avg_generation_tps": 201.66,
|
|
"avg_prompt_tps": 2519.75,
|
|
"evaluations": [
|
|
{
|
|
"prompt_id": "C01_code_repo_summary",
|
|
"content_preview": "",
|
|
"gen_tps": 201.61,
|
|
"prompt_tps": 1267.33
|
|
},
|
|
{
|
|
"prompt_id": "C02_security_audit_log",
|
|
"content_preview": "",
|
|
"gen_tps": 201.71,
|
|
"prompt_tps": 3772.17
|
|
}
|
|
],
|
|
"status": "OK"
|
|
},
|
|
"LFM2.5-2.6B_CPU": {
|
|
"name": "LFM2.5-2.6B",
|
|
"device": "CPU",
|
|
"ngl": 0,
|
|
"cold_start_sec": 4.83,
|
|
"process_vram_mib": 394,
|
|
"avg_generation_tps": 14.92,
|
|
"avg_prompt_tps": 396.87,
|
|
"evaluations": [
|
|
{
|
|
"prompt_id": "C01_code_repo_summary",
|
|
"content_preview": "",
|
|
"gen_tps": 15.68,
|
|
"prompt_tps": 394.56
|
|
},
|
|
{
|
|
"prompt_id": "C02_security_audit_log",
|
|
"content_preview": "",
|
|
"gen_tps": 14.16,
|
|
"prompt_tps": 399.17
|
|
}
|
|
],
|
|
"status": "OK"
|
|
}
|
|
},
|
|
"candidates_64k_vram": {
|
|
"Phi-4-14B": {
|
|
"name": "Phi-4-14B",
|
|
"status": "OK",
|
|
"cold_start_sec": 36.07,
|
|
"process_vram_mib": 15786,
|
|
"generation_tps": 59.38,
|
|
"prompt_tps": 189.01
|
|
},
|
|
"Qwen2.5-Coder-14B": {
|
|
"name": "Qwen2.5-Coder-14B",
|
|
"status": "OK",
|
|
"cold_start_sec": 48.07,
|
|
"process_vram_mib": 15400,
|
|
"generation_tps": 56.58,
|
|
"prompt_tps": 338.5
|
|
},
|
|
"Granite-4.2-8B": {
|
|
"name": "Granite-4.2-8B",
|
|
"status": "OK",
|
|
"cold_start_sec": 39.17,
|
|
"process_vram_mib": 11212,
|
|
"generation_tps": 82.35,
|
|
"prompt_tps": 152.05
|
|
},
|
|
"Qwen3-4B-2507": {
|
|
"name": "Qwen3-4B-2507",
|
|
"status": "OK",
|
|
"cold_start_sec": 4.51,
|
|
"process_vram_mib": 7976,
|
|
"generation_tps": 122.43,
|
|
"prompt_tps": 327.94
|
|
},
|
|
"Qwen3-Coder-30B-A3B": {
|
|
"name": "Qwen3-Coder-30B-A3B",
|
|
"status": "OK",
|
|
"cold_start_sec": 30.04,
|
|
"process_vram_mib": 21368,
|
|
"generation_tps": 107.5,
|
|
"prompt_tps": 115.08
|
|
}
|
|
},
|
|
"pair_coder_and_compressor": {
|
|
"status": "SUCCESS",
|
|
"name_a": "Qwen3-Coder-30B-A3B",
|
|
"name_b": "Qwen3-4B-2507",
|
|
"vram_a_mib": 21368,
|
|
"vram_b_mib": 5368,
|
|
"total_vram_mib": 26736,
|
|
"solo_a_tps": 110.08,
|
|
"solo_b_tps": 122.37,
|
|
"conc_a_tps": 54.23,
|
|
"conc_b_tps": 54.28,
|
|
"total_conc_tps": 108.5,
|
|
"ratio_vs_solo_a": 0.99
|
|
},
|
|
"pair_phi4_and_qwen25_14b": {
|
|
"status": "SUCCESS",
|
|
"name_a": "Phi-4-14B",
|
|
"name_b": "Qwen2.5-Coder-14B",
|
|
"vram_a_mib": 15786,
|
|
"vram_b_mib": 15400,
|
|
"total_vram_mib": 31186,
|
|
"solo_a_tps": 59.92,
|
|
"solo_b_tps": 56.86,
|
|
"conc_a_tps": 24.43,
|
|
"conc_b_tps": 24.45,
|
|
"total_conc_tps": 48.88,
|
|
"ratio_vs_solo_a": 0.82
|
|
},
|
|
"pair_qwen3moe_and_granite": {
|
|
"status": "SUCCESS",
|
|
"name_a": "Qwen3-Coder-30B-A3B",
|
|
"name_b": "Granite-4.2-8B",
|
|
"vram_a_mib": 19640,
|
|
"vram_b_mib": 8332,
|
|
"total_vram_mib": 27972,
|
|
"solo_a_tps": 109.31,
|
|
"solo_b_tps": 82.81,
|
|
"conc_a_tps": 42.73,
|
|
"conc_b_tps": 42.74,
|
|
"total_conc_tps": 85.47,
|
|
"ratio_vs_solo_a": 0.78
|
|
}
|
|
} |