Dataset Viewer
Auto-converted to Parquet Duplicate
accepted_at
timestamp[s]date
2026-08-19 00:00:00
2026-08-27 00:00:00
engine
dict
hardware
dict
id
stringlengths
36
103
links
dict
measured_at
stringlengths
10
24
metrics
dict
model
dict
model_instance
dict
recipe_id
stringlengths
30
124
sweep_id
stringlengths
36
103
null
{ "name": "vllm", "version": "0.26.0" }
{ "count": 1, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 96, "vendor": "nvidia" }
deepseek-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/deepseek", "recipe": "/api/v1/recipes/deepseek-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/deepseek-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep" }
2026-08-11
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 32768, "peak_decode_tok_s": 237.26, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "deepseek", "id": "deepseek", "name": "DeepSeek" }
{ "format": "safetensors", "id": "redhatai-deepseek-coder-v2-lite-instruct-fp8--fp8", "precision": "FP8", "repository": "RedHatAI/DeepSeek-Coder-V2-Lite-Instruct-FP8" }
deepseek-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp1
deepseek-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
2026-08-23T00:00:00
{ "name": "sparkinfer", "version": "NVIDIA vLLM 26.02 base" }
{ "count": 1, "id": "dgx-spark-gb10-128gb", "memory_gb": 128, "name": "NVIDIA DGX Spark GB10", "total_memory_gb": 128, "vendor": "nvidia" }
deepseek-v4-flash-0731-exl3-3bpw-dgxspark-sparkinfer-tp1-sweep
{ "hardware": "/api/v1/hardware/dgx-spark-gb10-128gb", "model": "/api/v1/models/deepseek-v4-flash-0731", "recipe": "/api/v1/recipes/deepseek-v4-flash-0731-exl3-3bpw-dgxspark-sparkinfer-tp1", "speed_sweep": "/api/v1/speed-sweeps/deepseek-v4-flash-0731-exl3-3bpw-dgxspark-sparkinfer-tp1-sweep" }
2026-08-21
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 370104, "peak_decode_tok_s": 47, "peak_decode_tok_s_per_stream": 47, "peak_prefill_tok_s": 630, "peak_vram_gb": null, "point_count": 2 }
{ "family": "deepseek", "id": "deepseek-v4-flash-0731", "name": "DeepSeek-V4-Flash-0731" }
{ "format": "EXL3", "id": "0xsero-deepseek-v4-flash-0731-spark--3-0-bpw-reap-k216", "precision": "3.0 bpw REAP-K216", "repository": "0xSero/deepseek-v4-flash-0731-spark" }
deepseek-v4-flash-0731-exl3-3bpw-dgxspark-sparkinfer-tp1
deepseek-v4-flash-0731-exl3-3bpw-dgxspark-sparkinfer-tp1-sweep
null
{ "name": "vllm", "version": "0.11.2.dev278+gccfeb3a8 (LimeChain sm86 fork)" }
{ "count": 8, "id": "rtx-3090-24gb", "memory_gb": 24, "name": "GeForce RTX 3090", "total_memory_gb": 192, "vendor": "nvidia" }
deepseek-v4-flash-0731-fp8-e4m3-rtx-3090-24gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-3090-24gb", "model": "/api/v1/models/deepseek-v4-flash-0731", "recipe": "/api/v1/recipes/deepseek-v4-flash-0731-fp8-e4m3-rtx-3090-24gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/deepseek-v4-flash-0731-fp8-e4m3-rtx-3090-24gb-vllm-tp1-sweep" }
2026-08-22
{ "best_ttft_ms": 267.1, "max_concurrency": 1, "max_context_tokens": 327680, "peak_decode_tok_s": 91.16, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 1344.17, "peak_vram_gb": 184.35, "point_count": 1 }
{ "family": "deepseek", "id": "deepseek-v4-flash-0731", "name": "DeepSeek-V4-Flash-0731" }
{ "format": "safetensors", "id": "deepseek-ai-deepseek-v4-flash-0731--fp8-e4m3", "precision": "FP8_E4M3", "repository": "deepseek-ai/DeepSeek-V4-Flash-0731" }
deepseek-v4-flash-0731-fp8-e4m3-rtx-3090-24gb-vllm-tp1
deepseek-v4-flash-0731-fp8-e4m3-rtx-3090-24gb-vllm-tp1-sweep
null
{ "name": "sglang", "version": "lmsysorg/sglang:dev-cu13-inkling-dspark@sha256:fbea1a4e25b26660dbc2384a27ead8817e9b7670f257b5c3143e0450d14524d7 (stock; SGLang commit b7252cc6b0c78b25ecea7ee5efa91a6ae37d0f19); no patches" }
{ "count": 2, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 192, "vendor": "nvidia" }
deepseek-v4-flash-0731-fp8-fp4-rtxpro6000-sglang-tp2-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/deepseek-v4-flash-0731", "recipe": "/api/v1/recipes/deepseek-v4-flash-0731-fp8-fp4-rtxpro6000-sglang-tp2", "speed_sweep": "/api/v1/speed-sweeps/deepseek-v4-flash-0731-fp8-fp4-rtxpro6000-sglang-tp2-sweep" }
2026-08-26
{ "best_ttft_ms": 80, "max_concurrency": 4, "max_context_tokens": 50, "peak_decode_tok_s": 155.2, "peak_decode_tok_s_per_stream": 57.85, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 5 }
{ "family": "deepseek", "id": "deepseek-v4-flash-0731", "name": "DeepSeek-V4-Flash-0731" }
{ "format": "safetensors", "id": "deepseek-ai-deepseek-v4-flash-0731--fp8", "precision": "FP8", "repository": "deepseek-ai/DeepSeek-V4-Flash-0731" }
deepseek-v4-flash-0731-fp8-fp4-rtxpro6000-sglang-tp2
deepseek-v4-flash-0731-fp8-fp4-rtxpro6000-sglang-tp2-sweep
null
{ "name": "vllm", "version": "LimeChain/vllm@ccfeb3a8b (release 4fd0bd1bf)" }
{ "count": 8, "id": "rtx-3090-24gb", "memory_gb": 24, "name": "GeForce RTX 3090", "total_memory_gb": 192, "vendor": "nvidia" }
deepseek-v4-flash-0731-fp8-rtx-3090-24gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-3090-24gb", "model": "/api/v1/models/deepseek-v4-flash-0731", "recipe": "/api/v1/recipes/deepseek-v4-flash-0731-fp8-rtx-3090-24gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/deepseek-v4-flash-0731-fp8-rtx-3090-24gb-vllm-tp1-sweep" }
2026-08-09
{ "best_ttft_ms": 362.046, "max_concurrency": 11, "max_context_tokens": 327680, "peak_decode_tok_s": 404.473815, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "deepseek", "id": "deepseek-v4-flash-0731", "name": "DeepSeek-V4-Flash-0731" }
{ "format": "safetensors", "id": "deepseek-ai-deepseek-v4-flash-0731--fp8", "precision": "FP8", "repository": "deepseek-ai/DeepSeek-V4-Flash-0731" }
deepseek-v4-flash-0731-fp8-rtx-3090-24gb-vllm-tp1
deepseek-v4-flash-0731-fp8-rtx-3090-24gb-vllm-tp1-sweep
null
{ "name": "lucebox", "version": "c994209" }
{ "count": 1, "id": "radeon-ai-pro-r9700-32gb", "memory_gb": 32, "name": "Radeon AI PRO R9700", "total_memory_gb": 32, "vendor": "amd" }
deepseek-v4-flash-0731-iq2-xxs-radeon-ai-pro-r9700-32gb-lucebox-tp1-7ibvc9zx-sweep
{ "hardware": "/api/v1/hardware/radeon-ai-pro-r9700-32gb", "model": "/api/v1/models/deepseek-v4-flash-0731", "recipe": "/api/v1/recipes/deepseek-v4-flash-0731-iq2-xxs-radeon-ai-pro-r9700-32gb-lucebox-tp1-7ibvc9zx", "speed_sweep": "/api/v1/speed-sweeps/deepseek-v4-flash-0731-iq2-xxs-radeon-ai-pro-r9700-32gb-luce...
2026-08-26T11:54:44.953Z
{ "best_ttft_ms": 2003.904, "max_concurrency": 1, "max_context_tokens": 8192, "peak_decode_tok_s": 59.432, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 196.117, "peak_vram_gb": 25.161, "point_count": 1 }
{ "family": "deepseek", "id": "deepseek-v4-flash-0731", "name": "DeepSeek-V4-Flash-0731" }
{ "format": "IQ2_XXS", "id": "deepseek-ai-deepseek-v4-flash-0731--iq2-xxs", "precision": "IQ2_XXS", "repository": "deepseek-ai/DeepSeek-V4-Flash-0731" }
deepseek-v4-flash-0731-iq2-xxs-radeon-ai-pro-r9700-32gb-lucebox-tp1-7ibvc9zx
deepseek-v4-flash-0731-iq2-xxs-radeon-ai-pro-r9700-32gb-lucebox-tp1-7ibvc9zx-sweep
2026-08-23T00:00:00
{ "name": "vllm", "version": "0.25.2+anemll-dspark-gx10-0.1.1" }
{ "count": 2, "id": "dgx-spark-gb10-128gb", "memory_gb": 128, "name": "NVIDIA DGX Spark GB10", "total_memory_gb": 256, "vendor": "nvidia" }
deepseek-v4-flash-0731-nvfp4-dgxspark-vllm-tp2-sweep
{ "hardware": "/api/v1/hardware/dgx-spark-gb10-128gb", "model": "/api/v1/models/deepseek-v4-flash-0731", "recipe": "/api/v1/recipes/deepseek-v4-flash-0731-nvfp4-dgxspark-vllm-tp2", "speed_sweep": "/api/v1/speed-sweeps/deepseek-v4-flash-0731-nvfp4-dgxspark-vllm-tp2-sweep" }
2026-08-01
{ "best_ttft_ms": null, "max_concurrency": 6, "max_context_tokens": 899994, "peak_decode_tok_s": 191.2, "peak_decode_tok_s_per_stream": 75.4, "peak_prefill_tok_s": 2563, "peak_vram_gb": null, "point_count": 19 }
{ "family": "deepseek", "id": "deepseek-v4-flash-0731", "name": "DeepSeek-V4-Flash-0731" }
{ "format": "ModelOpt", "id": "deepseek-ai-deepseek-v4-flash-0731--nvfp4", "precision": "NVFP4", "repository": "deepseek-ai/DeepSeek-V4-Flash-0731" }
deepseek-v4-flash-0731-nvfp4-dgxspark-vllm-tp2
deepseek-v4-flash-0731-nvfp4-dgxspark-vllm-tp2-sweep
null
{ "name": "vllm", "version": null }
{ "count": 4, "id": "rtx-5090-32gb", "memory_gb": 32, "name": "GeForce RTX 5090", "total_memory_gb": 128, "vendor": "nvidia" }
deepseek-v4-flash-2-bit-experts-vllm-moet-w2-fp8-kv-rtx-5090-32gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-5090-32gb", "model": "/api/v1/models/deepseek-v4-flash", "recipe": "/api/v1/recipes/deepseek-v4-flash-2-bit-experts-vllm-moet-w2-fp8-kv-rtx-5090-32gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/deepseek-v4-flash-2-bit-experts-vllm-moet-w2-fp8-kv-rtx-5090-32gb-vllm-tp1-swe...
2026-07-18
{ "best_ttft_ms": 55, "max_concurrency": 1, "max_context_tokens": 262144, "peak_decode_tok_s": 212, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 3900, "peak_vram_gb": 124, "point_count": 1 }
{ "family": "deepseek", "id": "deepseek-v4-flash", "name": "DeepSeek-V4-Flash" }
{ "format": "safetensors", "id": "deepseek-ai-deepseek-v4-flash--2-bit-experts-vllm-moet-w2-fp8-kv", "precision": "2-bit experts (vLLM-Moet W2) + FP8 KV", "repository": "deepseek-ai/DeepSeek-V4-Flash" }
deepseek-v4-flash-2-bit-experts-vllm-moet-w2-fp8-kv-rtx-5090-32gb-vllm-tp1
deepseek-v4-flash-2-bit-experts-vllm-moet-w2-fp8-kv-rtx-5090-32gb-vllm-tp1-sweep
null
{ "name": "mlx", "version": null }
{ "count": 1, "id": "apple-max-128gb", "memory_gb": 128, "name": "Apple Max 128GB, generation unspecified", "total_memory_gb": 128, "vendor": "apple" }
deepseek-v4-flash-2bit-dq-apple-max-128gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-max-128gb", "model": "/api/v1/models/deepseek-v4-flash", "recipe": "/api/v1/recipes/deepseek-v4-flash-2bit-dq-apple-max-128gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/deepseek-v4-flash-2bit-dq-apple-max-128gb-mlx-tp1-sweep" }
2026-04-27
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 17, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "deepseek", "id": "deepseek-v4-flash", "name": "DeepSeek-V4-Flash" }
{ "format": "MLX", "id": "mlx-community-deepseek-v4-flash-2bit-dq--2bit-dq", "precision": "2bit-DQ", "repository": "mlx-community/DeepSeek-V4-Flash-2bit-DQ" }
deepseek-v4-flash-2bit-dq-apple-max-128gb-mlx-tp1
deepseek-v4-flash-2bit-dq-apple-max-128gb-mlx-tp1-sweep
2026-08-23T00:00:00
{ "name": "vllm", "version": "0.25.2+anemll-dspark-gx10-0.1.1" }
{ "count": 2, "id": "dgx-spark-gb10-128gb", "memory_gb": 128, "name": "NVIDIA DGX Spark GB10", "total_memory_gb": 256, "vendor": "nvidia" }
deepseek-v4-flash-dspark-nvfp4-dgxspark-vllm-tp2-sweep
{ "hardware": "/api/v1/hardware/dgx-spark-gb10-128gb", "model": "/api/v1/models/deepseek-v4-flash-dspark", "recipe": "/api/v1/recipes/deepseek-v4-flash-dspark-nvfp4-dgxspark-vllm-tp2", "speed_sweep": "/api/v1/speed-sweeps/deepseek-v4-flash-dspark-nvfp4-dgxspark-vllm-tp2-sweep" }
2026-08-23
{ "best_ttft_ms": null, "max_concurrency": 6, "max_context_tokens": null, "peak_decode_tok_s": 153.7, "peak_decode_tok_s_per_stream": 66.6, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 6 }
{ "family": "deepseek", "id": "deepseek-v4-flash-dspark", "name": "DeepSeek-V4-Flash-DSpark" }
{ "format": "ModelOpt", "id": "deepseek-ai-deepseek-v4-flash-dspark--nvfp4", "precision": "NVFP4", "repository": "deepseek-ai/DeepSeek-V4-Flash-DSpark" }
deepseek-v4-flash-dspark-nvfp4-dgxspark-vllm-tp2
deepseek-v4-flash-dspark-nvfp4-dgxspark-vllm-tp2-sweep
null
{ "name": "vllm", "version": "0.1.dev1172+g4a6fd8747.xpu; local vLLM 264c7f2f7; XPU kernels 313156737; oneCCL 48fda4f0e" }
{ "count": 4, "id": "intel-arc-pro-b70-32gb", "memory_gb": 32, "name": "Intel Arc Pro B70", "total_memory_gb": 128, "vendor": "intel" }
deepseek-v4-flash-fp8-intel-arc-pro-b70-32gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/intel-arc-pro-b70-32gb", "model": "/api/v1/models/deepseek-v4-flash", "recipe": "/api/v1/recipes/deepseek-v4-flash-fp8-intel-arc-pro-b70-32gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/deepseek-v4-flash-fp8-intel-arc-pro-b70-32gb-vllm-tp1-sweep" }
2026-07-18
{ "best_ttft_ms": 342.7776570024434, "max_concurrency": 1, "max_context_tokens": 256, "peak_decode_tok_s": 80.82005189243556, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "deepseek", "id": "deepseek-v4-flash", "name": "DeepSeek-V4-Flash" }
{ "format": "safetensors", "id": "0xsero-deepseek-v4-flash-180b--fp8", "precision": "FP8", "repository": "0xSero/DeepSeek-V4-Flash-180B" }
deepseek-v4-flash-fp8-intel-arc-pro-b70-32gb-vllm-tp1
deepseek-v4-flash-fp8-intel-arc-pro-b70-32gb-vllm-tp1-sweep
2026-08-24T00:00:00
{ "name": "sglang", "version": "deepseek-v4-blackwell image; exact SGLang commit unreported" }
{ "count": 4, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 384, "vendor": "nvidia" }
deepseek-v4-flash-fp8-rtxpro6000-sglang-tp4-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/deepseek-v4-flash", "recipe": "/api/v1/recipes/deepseek-v4-flash-fp8-rtxpro6000-sglang-tp4", "speed_sweep": "/api/v1/speed-sweeps/deepseek-v4-flash-fp8-rtxpro6000-sglang-tp4-sweep" }
2026-08-24
{ "best_ttft_ms": 11337, "max_concurrency": 1, "max_context_tokens": 300000, "peak_decode_tok_s": 37.99, "peak_decode_tok_s_per_stream": 37.99, "peak_prefill_tok_s": 1273.1, "peak_vram_gb": null, "point_count": 7 }
{ "family": "deepseek", "id": "deepseek-v4-flash", "name": "DeepSeek-V4-Flash" }
{ "format": "safetensors", "id": "sgl-project-deepseek-v4-flash-fp8--fp8", "precision": "FP8", "repository": "sgl-project/DeepSeek-V4-Flash-FP8" }
deepseek-v4-flash-fp8-rtxpro6000-sglang-tp4
deepseek-v4-flash-fp8-rtxpro6000-sglang-tp4-sweep
null
{ "name": "mlx", "version": null }
{ "count": 1, "id": "apple-m3-ultra-512gb", "memory_gb": 512, "name": "Apple M3 Ultra 512GB", "total_memory_gb": 512, "vendor": "apple" }
deepseek-v4-flash-iq2xxs-apple-m3-ultra-512gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-m3-ultra-512gb", "model": "/api/v1/models/deepseek-v4-flash", "recipe": "/api/v1/recipes/deepseek-v4-flash-iq2xxs-apple-m3-ultra-512gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/deepseek-v4-flash-iq2xxs-apple-m3-ultra-512gb-mlx-tp1-sweep" }
2026-06-30
{ "best_ttft_ms": 300, "max_concurrency": 1, "max_context_tokens": 16384, "peak_decode_tok_s": 33, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "deepseek", "id": "deepseek-v4-flash", "name": "DeepSeek-V4-Flash" }
{ "format": "MLX", "id": "deepseek-ai-deepseek-v4-flash--iq2xxs", "precision": "IQ2XXS", "repository": "deepseek-ai/DeepSeek-V4-Flash" }
deepseek-v4-flash-iq2xxs-apple-m3-ultra-512gb-mlx-tp1
deepseek-v4-flash-iq2xxs-apple-m3-ultra-512gb-mlx-tp1-sweep
2026-08-24T00:00:00
{ "name": "vllm", "version": "0.1.dev17016+g27fd665bd.d20260526" }
{ "count": 1, "id": "dgx-spark-gb10-128gb", "memory_gb": 128, "name": "NVIDIA DGX Spark GB10", "total_memory_gb": 128, "vendor": "nvidia" }
deepseek-v4-flash-spark-200k-nvfp4-mxfp4-dgxspark-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/dgx-spark-gb10-128gb", "model": "/api/v1/models/deepseek-v4-flash-spark-200k", "recipe": "/api/v1/recipes/deepseek-v4-flash-spark-200k-nvfp4-mxfp4-dgxspark-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/deepseek-v4-flash-spark-200k-nvfp4-mxfp4-dgxspark-vllm-tp1-sweep" }
2026-05-27
{ "best_ttft_ms": 248216.9, "max_concurrency": 1, "max_context_tokens": 186390, "peak_decode_tok_s": 33.287, "peak_decode_tok_s_per_stream": 33.287, "peak_prefill_tok_s": 550.059, "peak_vram_gb": null, "point_count": 3 }
{ "family": "deepseek", "id": "deepseek-v4-flash-spark-200k", "name": "DeepSeek-V4-Flash-Spark-200K" }
{ "format": "safetensors", "id": "0xsero-deepseek-v4-flash-180b--nvfp4-mxfp4-experts", "precision": "NVFP4/MXFP4 experts", "repository": "0xSero/DeepSeek-V4-Flash-180B" }
deepseek-v4-flash-spark-200k-nvfp4-mxfp4-dgxspark-vllm-tp1
deepseek-v4-flash-spark-200k-nvfp4-mxfp4-dgxspark-vllm-tp1-sweep
null
{ "name": "vllm", "version": "0.1.dev1+ga275a28e0" }
{ "count": 1, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 96, "vendor": "nvidia" }
diffusiongemma-26b-a4b-bf16-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/diffusiongemma-26b-a4b", "recipe": "/api/v1/recipes/diffusiongemma-26b-a4b-bf16-rtx-pro-6000-blackwell-96gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/diffusiongemma-26b-a4b-bf16-rtx-pro-6000-blackwell-96gb-vllm-tp1-swee...
2026-07-02
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 8192, "peak_decode_tok_s": 528.7, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "gemma", "id": "diffusiongemma-26b-a4b", "name": "diffusiongemma-26B-A4B" }
{ "format": "safetensors", "id": "google-diffusiongemma-26b-a4b-it--bf16", "precision": "bf16", "repository": "google/diffusiongemma-26B-A4B-it" }
diffusiongemma-26b-a4b-bf16-rtx-pro-6000-blackwell-96gb-vllm-tp1
diffusiongemma-26b-a4b-bf16-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
null
{ "name": "vllm", "version": "0.22.1rc1.dev357+g74b5964f0" }
{ "count": 2, "id": "rtx-3090-24gb", "memory_gb": 24, "name": "GeForce RTX 3090", "total_memory_gb": 48, "vendor": "nvidia" }
diffusiongemma-26b-a4b-it-fp8-dynamic-rtx-3090-24gb-vllm-tp2-sweep
{ "hardware": "/api/v1/hardware/rtx-3090-24gb", "model": "/api/v1/models/diffusiongemma-26b-a4b-it", "recipe": "/api/v1/recipes/diffusiongemma-26b-a4b-it-fp8-dynamic-rtx-3090-24gb-vllm-tp2", "speed_sweep": "/api/v1/speed-sweeps/diffusiongemma-26b-a4b-it-fp8-dynamic-rtx-3090-24gb-vllm-tp2-sweep" }
2026-06-24
{ "best_ttft_ms": 1675.36, "max_concurrency": 1, "max_context_tokens": 262144, "peak_decode_tok_s": 177.55, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "gemma", "id": "diffusiongemma-26b-a4b-it", "name": "diffusiongemma-26B-A4B-it" }
{ "format": "safetensors", "id": "redhatai-diffusiongemma-26b-a4b-it-fp8-dynamic--fp8-dynamic", "precision": "FP8 dynamic", "repository": "RedHatAI/diffusiongemma-26B-A4B-it-FP8-dynamic" }
diffusiongemma-26b-a4b-it-fp8-dynamic-rtx-3090-24gb-vllm-tp2
diffusiongemma-26b-a4b-it-fp8-dynamic-rtx-3090-24gb-vllm-tp2-sweep
null
{ "name": "mlx", "version": "mlx-lm 0.31.3" }
{ "count": 1, "id": "apple-m5-max-128gb", "memory_gb": 128, "name": "Apple M5 Max 128GB", "total_memory_gb": 128, "vendor": "apple" }
gemma-3-4b-it-bf16-4bit-apple-m5-max-128gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-m5-max-128gb", "model": "/api/v1/models/gemma-3-4b-it-bf16", "recipe": "/api/v1/recipes/gemma-3-4b-it-bf16-4bit-apple-m5-max-128gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-3-4b-it-bf16-4bit-apple-m5-max-128gb-mlx-tp1-sweep" }
2026-06-22
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 175.116, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 3405.39, "peak_vram_gb": 3.016, "point_count": 1 }
{ "family": "gemma", "id": "gemma-3-4b-it-bf16", "name": "gemma-3-4b-it-bf16" }
{ "format": "MLX", "id": "mlx-community-gemma-3-text-4b-it-4bit--4bit", "precision": "4bit", "repository": "mlx-community/gemma-3-text-4b-it-4bit" }
gemma-3-4b-it-bf16-4bit-apple-m5-max-128gb-mlx-tp1
gemma-3-4b-it-bf16-4bit-apple-m5-max-128gb-mlx-tp1-sweep
null
{ "name": "mlx", "version": null }
{ "count": 1, "id": "apple-pro-64gb", "memory_gb": 64, "name": "Apple Pro 64GB, generation unspecified", "total_memory_gb": 64, "vendor": "apple" }
gemma-3-4b-it-bf16-4bit-apple-pro-64gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-pro-64gb", "model": "/api/v1/models/gemma-3-4b-it-bf16", "recipe": "/api/v1/recipes/gemma-3-4b-it-bf16-4bit-apple-pro-64gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-3-4b-it-bf16-4bit-apple-pro-64gb-mlx-tp1-sweep" }
2026-04-29
{ "best_ttft_ms": 109.3, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 105.032, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": 3.015, "point_count": 1 }
{ "family": "gemma", "id": "gemma-3-4b-it-bf16", "name": "gemma-3-4b-it-bf16" }
{ "format": "MLX", "id": "mlx-community-gemma-3-text-4b-it-4bit--4bit", "precision": "4bit", "repository": "mlx-community/gemma-3-text-4b-it-4bit" }
gemma-3-4b-it-bf16-4bit-apple-pro-64gb-mlx-tp1
gemma-3-4b-it-bf16-4bit-apple-pro-64gb-mlx-tp1-sweep
null
{ "name": "vllm", "version": null }
{ "count": 4, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 384, "vendor": "nvidia" }
gemma-4-12b-bf16-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/gemma-4-12b", "recipe": "/api/v1/recipes/gemma-4-12b-bf16-rtx-pro-6000-blackwell-96gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-12b-bf16-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep" }
2026-06-30
{ "best_ttft_ms": 34.69645799850696, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 145.6499652268921, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 11255.63223431487, "peak_vram_gb": null, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-12b", "name": "gemma-4-12B" }
{ "format": "safetensors", "id": "google-gemma-4-12b-it--bf16", "precision": "BF16", "repository": "google/gemma-4-12B-it" }
gemma-4-12b-bf16-rtx-pro-6000-blackwell-96gb-vllm-tp1
gemma-4-12b-bf16-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
null
{ "name": "vllm", "version": null }
{ "count": 1, "id": "rtx-5090-32gb", "memory_gb": 32, "name": "GeForce RTX 5090", "total_memory_gb": 32, "vendor": "nvidia" }
gemma-4-12b-coder-fable5-composer2-5-v1-gguf-nvfp4-rtx-5090-32gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-5090-32gb", "model": "/api/v1/models/gemma-4-12b-coder-fable5-composer2-5-v1-gguf", "recipe": "/api/v1/recipes/gemma-4-12b-coder-fable5-composer2-5-v1-gguf-nvfp4-rtx-5090-32gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-12b-coder-fable5-composer2-5-v1-gguf-nvfp4-r...
2026-07-23
{ "best_ttft_ms": 31.85, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 183, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 1067.6, "peak_vram_gb": null, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-12b-coder-fable5-composer2-5-v1-gguf", "name": "gemma-4-12B-coder-fable5-composer2.5-v1-GGUF" }
{ "format": "safetensors", "id": "sakamakismile-gemma-4-12b-coder-fable5-composer2-5-mtp-nvfp4--nvfp4", "precision": "NVFP4", "repository": "sakamakismile/gemma-4-12B-coder-fable5-composer2.5-MTP-NVFP4" }
gemma-4-12b-coder-fable5-composer2-5-v1-gguf-nvfp4-rtx-5090-32gb-vllm-tp1
gemma-4-12b-coder-fable5-composer2-5-v1-gguf-nvfp4-rtx-5090-32gb-vllm-tp1-sweep
2026-08-25T00:00:00
{ "name": "tabbyapi", "version": "0.0.1+e632af41" }
{ "count": 1, "id": "rtx-2000-ada-16gb", "memory_gb": 16, "name": "NVIDIA RTX 2000 Ada Generation", "total_memory_gb": 16, "vendor": "nvidia" }
gemma-4-12b-it-exl3-3bpw-rtx2000ada-tabbyapi-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-2000-ada-16gb", "model": "/api/v1/models/gemma-4-12b-it", "recipe": "/api/v1/recipes/gemma-4-12b-it-exl3-3bpw-rtx2000ada-tabbyapi-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-12b-it-exl3-3bpw-rtx2000ada-tabbyapi-tp1-sweep" }
2026-08-25
{ "best_ttft_ms": 311678.8, "max_concurrency": 4, "max_context_tokens": 130560, "peak_decode_tok_s": null, "peak_decode_tok_s_per_stream": 20.03, "peak_prefill_tok_s": 418.89, "peak_vram_gb": null, "point_count": 3 }
{ "family": "gemma", "id": "gemma-4-12b-it", "name": "Gemma-4-12B-it" }
{ "format": "EXL3", "id": "turboderp-gemma-4-12b-it-exl3--3-bpw", "precision": "3 bpw", "repository": "turboderp/gemma-4-12B-it-exl3" }
gemma-4-12b-it-exl3-3bpw-rtx2000ada-tabbyapi-tp1
gemma-4-12b-it-exl3-3bpw-rtx2000ada-tabbyapi-tp1-sweep
2026-08-24T00:00:00
{ "name": "tabbyapi", "version": "0.0.1+e632af41" }
{ "count": 1, "id": "rtx-3090-24gb", "memory_gb": 24, "name": "GeForce RTX 3090", "total_memory_gb": 24, "vendor": "nvidia" }
gemma-4-12b-it-exl3-4bpw-rtx3090-tabbyapi-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-3090-24gb", "model": "/api/v1/models/gemma-4-12b-it", "recipe": "/api/v1/recipes/gemma-4-12b-it-exl3-4bpw-rtx3090-tabbyapi-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-12b-it-exl3-4bpw-rtx3090-tabbyapi-tp1-sweep" }
2026-08-24
{ "best_ttft_ms": 133291.9, "max_concurrency": 4, "max_context_tokens": 130560, "peak_decode_tok_s": null, "peak_decode_tok_s_per_stream": 26.56, "peak_prefill_tok_s": 979.93, "peak_vram_gb": null, "point_count": 3 }
{ "family": "gemma", "id": "gemma-4-12b-it", "name": "Gemma-4-12B-it" }
{ "format": "EXL3", "id": "turboderp-gemma-4-12b-it-exl3--4-bpw", "precision": "4 bpw", "repository": "turboderp/gemma-4-12B-it-exl3" }
gemma-4-12b-it-exl3-4bpw-rtx3090-tabbyapi-tp1
gemma-4-12b-it-exl3-4bpw-rtx3090-tabbyapi-tp1-sweep
2026-08-24T00:00:00
{ "name": "tabbyapi", "version": "0.0.1+e632af41" }
{ "count": 1, "id": "rtx-4090-24gb", "memory_gb": 24, "name": "GeForce RTX 4090", "total_memory_gb": 24, "vendor": "nvidia" }
gemma-4-12b-it-exl3-4bpw-rtx4090-tabbyapi-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-4090-24gb", "model": "/api/v1/models/gemma-4-12b-it", "recipe": "/api/v1/recipes/gemma-4-12b-it-exl3-4bpw-rtx4090-tabbyapi-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-12b-it-exl3-4bpw-rtx4090-tabbyapi-tp1-sweep" }
2026-08-24
{ "best_ttft_ms": 59240.8, "max_concurrency": 4, "max_context_tokens": 130560, "peak_decode_tok_s": null, "peak_decode_tok_s_per_stream": 74.74, "peak_prefill_tok_s": 2204.9, "peak_vram_gb": null, "point_count": 3 }
{ "family": "gemma", "id": "gemma-4-12b-it", "name": "Gemma-4-12B-it" }
{ "format": "EXL3", "id": "turboderp-gemma-4-12b-it-exl3--4-bpw", "precision": "4 bpw", "repository": "turboderp/gemma-4-12B-it-exl3" }
gemma-4-12b-it-exl3-4bpw-rtx4090-tabbyapi-tp1
gemma-4-12b-it-exl3-4bpw-rtx4090-tabbyapi-tp1-sweep
2026-08-25T00:00:00
{ "name": "tabbyapi", "version": "0.0.1+e632af41" }
{ "count": 1, "id": "rtx-pro-4000-blackwell-24gb", "memory_gb": 24, "name": "NVIDIA RTX PRO 4000 Blackwell", "total_memory_gb": 24, "vendor": "nvidia" }
gemma-4-12b-it-exl3-4bpw-rtxpro4000-tabbyapi-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-4000-blackwell-24gb", "model": "/api/v1/models/gemma-4-12b-it", "recipe": "/api/v1/recipes/gemma-4-12b-it-exl3-4bpw-rtxpro4000-tabbyapi-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-12b-it-exl3-4bpw-rtxpro4000-tabbyapi-tp1-sweep" }
2026-08-24
{ "best_ttft_ms": 119956.7, "max_concurrency": 4, "max_context_tokens": 130560, "peak_decode_tok_s": null, "peak_decode_tok_s_per_stream": 37.27, "peak_prefill_tok_s": 1089.31, "peak_vram_gb": null, "point_count": 3 }
{ "family": "gemma", "id": "gemma-4-12b-it", "name": "Gemma-4-12B-it" }
{ "format": "EXL3", "id": "turboderp-gemma-4-12b-it-exl3--4-bpw", "precision": "4 bpw", "repository": "turboderp/gemma-4-12B-it-exl3" }
gemma-4-12b-it-exl3-4bpw-rtxpro4000-tabbyapi-tp1
gemma-4-12b-it-exl3-4bpw-rtxpro4000-tabbyapi-tp1-sweep
null
{ "name": "vllm", "version": "0.20.2rc1.dev2+gc51df4300.d20260523" }
{ "count": 4, "id": "intel-arc-pro-b70-32gb", "memory_gb": 32, "name": "Intel Arc Pro B70", "total_memory_gb": 128, "vendor": "intel" }
gemma-4-12b-it-int4-autoround-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-sweep
{ "hardware": "/api/v1/hardware/intel-arc-pro-b70-32gb", "model": "/api/v1/models/gemma-4-12b-it", "recipe": "/api/v1/recipes/gemma-4-12b-it-int4-autoround-w4a16-intel-arc-pro-b70-32gb-vllm-tp4", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-12b-it-int4-autoround-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-sweep" }
2026-06-07
{ "best_ttft_ms": 2529.805063267122, "max_concurrency": 8, "max_context_tokens": 32768, "peak_decode_tok_s": 796.1763558639424, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-12b-it", "name": "Gemma-4-12B-it" }
{ "format": "safetensors", "id": "intel-gemma-4-12b-it-int4-autoround--int4-autoround-w4a16", "precision": "INT4 AutoRound W4A16", "repository": "Intel/gemma-4-12B-it-int4-AutoRound" }
gemma-4-12b-it-int4-autoround-w4a16-intel-arc-pro-b70-32gb-vllm-tp4
gemma-4-12b-it-int4-autoround-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-sweep
2026-08-23T00:00:00
{ "name": "sglang", "version": "0.0.0.dev1+geec794bce" }
{ "count": 1, "id": "rtx-5090-32gb", "memory_gb": 32, "name": "GeForce RTX 5090", "total_memory_gb": 32, "vendor": "nvidia" }
gemma-4-12b-it-nvfp4-rtx5090-sglang-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-5090-32gb", "model": "/api/v1/models/gemma-4-12b-it", "recipe": "/api/v1/recipes/gemma-4-12b-it-nvfp4-rtx5090-sglang-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-12b-it-nvfp4-rtx5090-sglang-tp1-sweep" }
2026-08-23
{ "best_ttft_ms": 244.6, "max_concurrency": 1, "max_context_tokens": 130560, "peak_decode_tok_s": null, "peak_decode_tok_s_per_stream": 39.94, "peak_prefill_tok_s": 532526.53, "peak_vram_gb": null, "point_count": 2 }
{ "family": "gemma", "id": "gemma-4-12b-it", "name": "Gemma-4-12B-it" }
{ "format": "ModelOpt", "id": "unsloth-gemma-4-12b-it-nvfp4--nvfp4", "precision": "NVFP4", "repository": "unsloth/gemma-4-12b-it-NVFP4" }
gemma-4-12b-it-nvfp4-rtx5090-sglang-tp1
gemma-4-12b-it-nvfp4-rtx5090-sglang-tp1-sweep
2026-08-24T00:00:00
{ "name": "sglang", "version": "0.0.0.dev1+geec794bce" }
{ "count": 1, "id": "rtx-pro-4500-blackwell-32gb", "memory_gb": 32, "name": "NVIDIA RTX PRO 4500 Blackwell", "total_memory_gb": 32, "vendor": "nvidia" }
gemma-4-12b-it-nvfp4-rtxpro4500-sglang-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-4500-blackwell-32gb", "model": "/api/v1/models/gemma-4-12b-it", "recipe": "/api/v1/recipes/gemma-4-12b-it-nvfp4-rtxpro4500-sglang-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-12b-it-nvfp4-rtxpro4500-sglang-tp1-sweep" }
2026-08-24
{ "best_ttft_ms": 263.2, "max_concurrency": 1, "max_context_tokens": 130560, "peak_decode_tok_s": null, "peak_decode_tok_s_per_stream": 38.23, "peak_prefill_tok_s": 502159.21, "peak_vram_gb": null, "point_count": 2 }
{ "family": "gemma", "id": "gemma-4-12b-it", "name": "Gemma-4-12B-it" }
{ "format": "ModelOpt", "id": "unsloth-gemma-4-12b-it-nvfp4--nvfp4", "precision": "NVFP4", "repository": "unsloth/gemma-4-12b-it-NVFP4" }
gemma-4-12b-it-nvfp4-rtxpro4500-sglang-tp1
gemma-4-12b-it-nvfp4-rtxpro4500-sglang-tp1-sweep
2026-08-24T00:00:00
{ "name": "sglang", "version": "0.0.0.dev1+geec794bce" }
{ "count": 1, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 96, "vendor": "nvidia" }
gemma-4-12b-it-nvfp4-rtxpro6000-sglang-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/gemma-4-12b-it", "recipe": "/api/v1/recipes/gemma-4-12b-it-nvfp4-rtxpro6000-sglang-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-12b-it-nvfp4-rtxpro6000-sglang-tp1-sweep" }
2026-08-24
{ "best_ttft_ms": 258.3, "max_concurrency": 4, "max_context_tokens": 130560, "peak_decode_tok_s": null, "peak_decode_tok_s_per_stream": 38.3, "peak_prefill_tok_s": 505439.9, "peak_vram_gb": null, "point_count": 6 }
{ "family": "gemma", "id": "gemma-4-12b-it", "name": "Gemma-4-12B-it" }
{ "format": "ModelOpt", "id": "unsloth-gemma-4-12b-it-nvfp4--nvfp4", "precision": "NVFP4", "repository": "unsloth/gemma-4-12b-it-NVFP4" }
gemma-4-12b-it-nvfp4-rtxpro6000-sglang-tp1
gemma-4-12b-it-nvfp4-rtxpro6000-sglang-tp1-sweep
null
{ "name": "vllm", "version": null }
{ "count": 1, "id": "rtx-4070-12gb", "memory_gb": 12, "name": "GeForce RTX 4070", "total_memory_gb": 12, "vendor": "nvidia" }
gemma-4-12b-it-q4-k-m-rtx-4070-12gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-4070-12gb", "model": "/api/v1/models/gemma-4-12b-it", "recipe": "/api/v1/recipes/gemma-4-12b-it-q4-k-m-rtx-4070-12gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-12b-it-q4-k-m-rtx-4070-12gb-vllm-tp1-sweep" }
2026-07-20
{ "best_ttft_ms": 153.69, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 48.9, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 240.7, "peak_vram_gb": null, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-12b-it", "name": "Gemma-4-12B-it" }
{ "format": "safetensors", "id": "yuxinlu1-gemma-4-12b-agentic-fable5-composer2-5-v2-3-5x-tau2-gguf--q4-k-m", "precision": "Q4_K_M", "repository": "yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF" }
gemma-4-12b-it-q4-k-m-rtx-4070-12gb-vllm-tp1
gemma-4-12b-it-q4-k-m-rtx-4070-12gb-vllm-tp1-sweep
null
{ "name": "vllm", "version": null }
{ "count": 1, "id": "rx-7700-xt-12gb", "memory_gb": 12, "name": "AMD Radeon RX 7700 XT", "total_memory_gb": 12, "vendor": "amd" }
gemma-4-12b-it-qat-q4-0-unquantized-q4-0-rx-7700-xt-12gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rx-7700-xt-12gb", "model": "/api/v1/models/gemma-4-12b-it-qat-q4-0-unquantized", "recipe": "/api/v1/recipes/gemma-4-12b-it-qat-q4-0-unquantized-q4-0-rx-7700-xt-12gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-12b-it-qat-q4-0-unquantized-q4-0-rx-7700-xt-12gb-vllm-tp1-s...
2026-07-25
{ "best_ttft_ms": 225.68, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 40.4, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 163.9, "peak_vram_gb": null, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-12b-it-qat-q4-0-unquantized", "name": "gemma-4-12B-it-qat-q4_0-unquantized" }
{ "format": "safetensors", "id": "google-gemma-4-12b-it-qat-q4-0-gguf--q4-0", "precision": "Q4_0", "repository": "google/gemma-4-12B-it-qat-q4_0-gguf" }
gemma-4-12b-it-qat-q4-0-unquantized-q4-0-rx-7700-xt-12gb-vllm-tp1
gemma-4-12b-it-qat-q4-0-unquantized-q4-0-rx-7700-xt-12gb-vllm-tp1-sweep
null
{ "name": "vllm", "version": null }
{ "count": 1, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 96, "vendor": "nvidia" }
gemma-4-26b-a4b-int8-per-channel-weight-only-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/gemma-4-26b-a4b", "recipe": "/api/v1/recipes/gemma-4-26b-a4b-int8-per-channel-weight-only-rtx-pro-6000-blackwell-96gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-26b-a4b-int8-per-channel-weight-only-rtx-pro-6000-b...
2026-07-02
{ "best_ttft_ms": 27.69, "max_concurrency": 1, "max_context_tokens": 8192, "peak_decode_tok_s": 276.7, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 1228, "peak_vram_gb": null, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-26b-a4b", "name": "gemma-4-26B-A4B" }
{ "format": "safetensors", "id": "google-gemma-4-26b-a4b-it--int8-per-channel-weight-only", "precision": "int8_per_channel_weight_only", "repository": "google/gemma-4-26B-A4B-it" }
gemma-4-26b-a4b-int8-per-channel-weight-only-rtx-pro-6000-blackwell-96gb-vllm-tp1
gemma-4-26b-a4b-int8-per-channel-weight-only-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
null
{ "name": "vllm", "version": null }
{ "count": 1, "id": "rtx-5090-32gb", "memory_gb": 32, "name": "GeForce RTX 5090", "total_memory_gb": 32, "vendor": "nvidia" }
gemma-4-26b-a4b-int8-rtx-5090-32gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-5090-32gb", "model": "/api/v1/models/gemma-4-26b-a4b", "recipe": "/api/v1/recipes/gemma-4-26b-a4b-int8-rtx-5090-32gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-26b-a4b-int8-rtx-5090-32gb-vllm-tp1-sweep" }
2026-07-23
{ "best_ttft_ms": 30.97, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 296.4, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 1097.8, "peak_vram_gb": null, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-26b-a4b", "name": "gemma-4-26B-A4B" }
{ "format": "safetensors", "id": "google-gemma-4-26b-a4b-it--int8", "precision": "INT8", "repository": "google/gemma-4-26B-A4B-it" }
gemma-4-26b-a4b-int8-rtx-5090-32gb-vllm-tp1
gemma-4-26b-a4b-int8-rtx-5090-32gb-vllm-tp1-sweep
null
{ "name": "vllm", "version": "0.26.0" }
{ "count": 2, "id": "rtx-3090-24gb", "memory_gb": 24, "name": "GeForce RTX 3090", "total_memory_gb": 48, "vendor": "nvidia" }
gemma-4-26b-a4b-int8-w8a16-rtx-3090-24gb-vllm-tp2-sweep
{ "hardware": "/api/v1/hardware/rtx-3090-24gb", "model": "/api/v1/models/gemma-4-26b-a4b", "recipe": "/api/v1/recipes/gemma-4-26b-a4b-int8-w8a16-rtx-3090-24gb-vllm-tp2", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-26b-a4b-int8-w8a16-rtx-3090-24gb-vllm-tp2-sweep" }
2026-07-25
{ "best_ttft_ms": 46.5, "max_concurrency": 1, "max_context_tokens": 262144, "peak_decode_tok_s": 162.9, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 8865, "peak_vram_gb": 45.4, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-26b-a4b", "name": "gemma-4-26B-A4B" }
{ "format": "safetensors", "id": "google-gemma-4-26b-a4b-it--int8-w8a16", "precision": "INT8_W8A16", "repository": "google/gemma-4-26B-A4B-it" }
gemma-4-26b-a4b-int8-w8a16-rtx-3090-24gb-vllm-tp2
gemma-4-26b-a4b-int8-w8a16-rtx-3090-24gb-vllm-tp2-sweep
null
{ "name": "mlx", "version": "0.31.2" }
{ "count": 1, "id": "apple-m3-ultra-512gb", "memory_gb": 512, "name": "Apple M3 Ultra 512GB", "total_memory_gb": 512, "vendor": "apple" }
gemma-4-26b-a4b-it-4bit-apple-m3-ultra-512gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-m3-ultra-512gb", "model": "/api/v1/models/gemma-4-26b-a4b-it", "recipe": "/api/v1/recipes/gemma-4-26b-a4b-it-4bit-apple-m3-ultra-512gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-26b-a4b-it-4bit-apple-m3-ultra-512gb-mlx-tp1-sweep" }
2026-04-26
{ "best_ttft_ms": 145, "max_concurrency": 1, "max_context_tokens": 8192, "peak_decode_tok_s": 96.71, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-26b-a4b-it", "name": "gemma-4-26B-A4B-it" }
{ "format": "MLX", "id": "mlx-community-gemma-4-26b-a4b-it-4bit--4bit", "precision": "4bit", "repository": "mlx-community/gemma-4-26b-a4b-it-4bit" }
gemma-4-26b-a4b-it-4bit-apple-m3-ultra-512gb-mlx-tp1
gemma-4-26b-a4b-it-4bit-apple-m3-ultra-512gb-mlx-tp1-sweep
null
{ "name": "mlx", "version": null }
{ "count": 1, "id": "apple-max-128gb", "memory_gb": 128, "name": "Apple Max 128GB, generation unspecified", "total_memory_gb": 128, "vendor": "apple" }
gemma-4-26b-a4b-it-4bit-apple-max-128gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-max-128gb", "model": "/api/v1/models/gemma-4-26b-a4b-it", "recipe": "/api/v1/recipes/gemma-4-26b-a4b-it-4bit-apple-max-128gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-26b-a4b-it-4bit-apple-max-128gb-mlx-tp1-sweep" }
2026-04-26
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 117, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-26b-a4b-it", "name": "gemma-4-26B-A4B-it" }
{ "format": "MLX", "id": "mlx-community-gemma-4-26b-a4b-it-4bit--4bit", "precision": "4bit", "repository": "mlx-community/gemma-4-26b-a4b-it-4bit" }
gemma-4-26b-a4b-it-4bit-apple-max-128gb-mlx-tp1
gemma-4-26b-a4b-it-4bit-apple-max-128gb-mlx-tp1-sweep
null
{ "name": "vllm", "version": null }
{ "count": 1, "id": "rtx-5090-32gb", "memory_gb": 32, "name": "GeForce RTX 5090", "total_memory_gb": 32, "vendor": "nvidia" }
gemma-4-26b-a4b-it-int4-rtx-5090-32gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-5090-32gb", "model": "/api/v1/models/gemma-4-26b-a4b-it", "recipe": "/api/v1/recipes/gemma-4-26b-a4b-it-int4-rtx-5090-32gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-26b-a4b-it-int4-rtx-5090-32gb-vllm-tp1-sweep" }
2026-07-23
{ "best_ttft_ms": 28.85, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 378.1, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 1178.3, "peak_vram_gb": null, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-26b-a4b-it", "name": "gemma-4-26B-A4B-it" }
{ "format": "safetensors", "id": "cyankiwi-gemma-4-26b-a4b-it-awq-4bit--int4", "precision": "INT4", "repository": "cyankiwi/gemma-4-26B-A4B-it-AWQ-4bit" }
gemma-4-26b-a4b-it-int4-rtx-5090-32gb-vllm-tp1
gemma-4-26b-a4b-it-int4-rtx-5090-32gb-vllm-tp1-sweep
null
{ "name": "mlx", "version": "0.31.3" }
{ "count": 1, "id": "apple-m4-pro-48gb", "memory_gb": 48, "name": "Apple M4 Pro 48GB", "total_memory_gb": 48, "vendor": "apple" }
gemma-4-26b-a4b-it-mlx-4bit-apple-m4-pro-48gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-m4-pro-48gb", "model": "/api/v1/models/gemma-4-26b-a4b-it", "recipe": "/api/v1/recipes/gemma-4-26b-a4b-it-mlx-4bit-apple-m4-pro-48gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-26b-a4b-it-mlx-4bit-apple-m4-pro-48gb-mlx-tp1-sweep" }
2026-06-27
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 2176, "peak_decode_tok_s": 65.346, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 782.941, "peak_vram_gb": 15.74, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-26b-a4b-it", "name": "gemma-4-26B-A4B-it" }
{ "format": "MLX", "id": "lmstudio-community-gemma-4-26b-a4b-it-qat-mlx-4bit--mlx-4bit", "precision": "MLX-4bit", "repository": "lmstudio-community/gemma-4-26B-A4B-it-QAT-MLX-4bit" }
gemma-4-26b-a4b-it-mlx-4bit-apple-m4-pro-48gb-mlx-tp1
gemma-4-26b-a4b-it-mlx-4bit-apple-m4-pro-48gb-mlx-tp1-sweep
null
{ "name": "vllm", "version": "0.20.1" }
{ "count": 1, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 96, "vendor": "nvidia" }
gemma-4-26b-a4b-it-nvfp4-modelopt-fp4-w4a4-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/gemma-4-26b-a4b-it", "recipe": "/api/v1/recipes/gemma-4-26b-a4b-it-nvfp4-modelopt-fp4-w4a4-rtx-pro-6000-blackwell-96gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-26b-a4b-it-nvfp4-modelopt-fp4-w4a4-rtx-pro-6000-bl...
2026-05-10
{ "best_ttft_ms": null, "max_concurrency": 16, "max_context_tokens": 65536, "peak_decode_tok_s": 1048, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-26b-a4b-it", "name": "gemma-4-26B-A4B-it" }
{ "format": "safetensors", "id": "nvidia-gemma-4-26b-a4b-nvfp4--nvfp4-modelopt-fp4-w4a4", "precision": "NVFP4 (modelopt_fp4 W4A4)", "repository": "nvidia/Gemma-4-26B-A4B-NVFP4" }
gemma-4-26b-a4b-it-nvfp4-modelopt-fp4-w4a4-rtx-pro-6000-blackwell-96gb-vllm-tp1
gemma-4-26b-a4b-it-nvfp4-modelopt-fp4-w4a4-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
null
{ "name": "vllm", "version": "0.20.1" }
{ "count": 1, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 96, "vendor": "nvidia" }
gemma-4-26b-a4b-it-nvfp4-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/gemma-4-26b-a4b-it", "recipe": "/api/v1/recipes/gemma-4-26b-a4b-it-nvfp4-rtx-pro-6000-blackwell-96gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-26b-a4b-it-nvfp4-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep" }
2026-05-10
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 65536, "peak_decode_tok_s": 128.3, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": 87.5, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-26b-a4b-it", "name": "gemma-4-26B-A4B-it" }
{ "format": "safetensors", "id": "nvidia-gemma-4-26b-a4b-nvfp4--nvfp4", "precision": "NVFP4", "repository": "nvidia/Gemma-4-26B-A4B-NVFP4" }
gemma-4-26b-a4b-it-nvfp4-rtx-pro-6000-blackwell-96gb-vllm-tp1
gemma-4-26b-a4b-it-nvfp4-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
null
{ "name": "vllm", "version": null }
{ "count": 1, "id": "rtx-4070-12gb", "memory_gb": 12, "name": "GeForce RTX 4070", "total_memory_gb": 12, "vendor": "nvidia" }
gemma-4-26b-a4b-q4-k-xl-rtx-4070-12gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-4070-12gb", "model": "/api/v1/models/gemma-4-26b-a4b", "recipe": "/api/v1/recipes/gemma-4-26b-a4b-q4-k-xl-rtx-4070-12gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-26b-a4b-q4-k-xl-rtx-4070-12gb-vllm-tp1-sweep" }
2026-07-20
{ "best_ttft_ms": 745.12, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 30.6, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 88024.2, "peak_vram_gb": null, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-26b-a4b", "name": "gemma-4-26B-A4B" }
{ "format": "safetensors", "id": "google-gemma-4-26b-a4b-it--q4-k-xl", "precision": "Q4_K_XL", "repository": "google/gemma-4-26B-A4B-it" }
gemma-4-26b-a4b-q4-k-xl-rtx-4070-12gb-vllm-tp1
gemma-4-26b-a4b-q4-k-xl-rtx-4070-12gb-vllm-tp1-sweep
null
{ "name": "vllm", "version": null }
{ "count": 4, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 384, "vendor": "nvidia" }
gemma-4-31b-bf16-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/gemma-4-31b", "recipe": "/api/v1/recipes/gemma-4-31b-bf16-rtx-pro-6000-blackwell-96gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-31b-bf16-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep" }
2026-06-30
{ "best_ttft_ms": 42.98081950037158, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 98.19400282700367, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 5985.067597638829, "peak_vram_gb": null, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-31b", "name": "gemma-4-31B" }
{ "format": "safetensors", "id": "google-gemma-4-31b-it--bf16", "precision": "BF16", "repository": "google/gemma-4-31B-it" }
gemma-4-31b-bf16-rtx-pro-6000-blackwell-96gb-vllm-tp1
gemma-4-31b-bf16-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
null
{ "name": "vllm", "version": null }
{ "count": 1, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 96, "vendor": "nvidia" }
gemma-4-31b-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/gemma-4-31b", "recipe": "/api/v1/recipes/gemma-4-31b-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-31b-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep" }
2026-05-06
{ "best_ttft_ms": 68, "max_concurrency": 1, "max_context_tokens": 131072, "peak_decode_tok_s": 58.6, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 4125, "peak_vram_gb": null, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-31b", "name": "gemma-4-31B" }
{ "format": "safetensors", "id": "google-gemma-4-31b-it--fp8", "precision": "FP8", "repository": "google/gemma-4-31B-it" }
gemma-4-31b-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp1
gemma-4-31b-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
null
{ "name": "vllm", "version": null }
{ "count": 2, "id": "rtx-3090-24gb", "memory_gb": 24, "name": "GeForce RTX 3090", "total_memory_gb": 48, "vendor": "nvidia" }
gemma-4-31b-int4-rtx-3090-24gb-vllm-tp2-sweep
{ "hardware": "/api/v1/hardware/rtx-3090-24gb", "model": "/api/v1/models/gemma-4-31b", "recipe": "/api/v1/recipes/gemma-4-31b-int4-rtx-3090-24gb-vllm-tp2", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-31b-int4-rtx-3090-24gb-vllm-tp2-sweep" }
2026-06-27
{ "best_ttft_ms": 71, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 135.12, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": 45.1, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-31b", "name": "gemma-4-31B" }
{ "format": "safetensors", "id": "google-gemma-4-31b-it--int4", "precision": "int4", "repository": "google/gemma-4-31B-it" }
gemma-4-31b-int4-rtx-3090-24gb-vllm-tp2
gemma-4-31b-int4-rtx-3090-24gb-vllm-tp2-sweep
null
{ "name": "vllm", "version": "0.20.2rc1.dev93+g51f22dcfd" }
{ "count": 2, "id": "rtx-3090-24gb", "memory_gb": 24, "name": "GeForce RTX 3090", "total_memory_gb": 48, "vendor": "nvidia" }
gemma-4-31b-int8-rtx-3090-24gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-3090-24gb", "model": "/api/v1/models/gemma-4-31b", "recipe": "/api/v1/recipes/gemma-4-31b-int8-rtx-3090-24gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-31b-int8-rtx-3090-24gb-vllm-tp1-sweep" }
2026-05-07
{ "best_ttft_ms": 110.23, "max_concurrency": 1, "max_context_tokens": 8192, "peak_decode_tok_s": 87.67, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": 21.6, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-31b", "name": "gemma-4-31B" }
{ "format": "safetensors", "id": "google-gemma-4-31b-it--int8", "precision": "int8", "repository": "google/gemma-4-31B-it" }
gemma-4-31b-int8-rtx-3090-24gb-vllm-tp1
gemma-4-31b-int8-rtx-3090-24gb-vllm-tp1-sweep
null
{ "name": "mlx", "version": "mlx-vlm 0.6.3 / mlx 0.32.0.dev20260702+e9463bb (qmv_wide)" }
{ "count": 1, "id": "apple-m5-pro-48gb", "memory_gb": 48, "name": "Apple M5 Pro 48GB", "total_memory_gb": 48, "vendor": "apple" }
gemma-4-31b-v2-int4-apple-m5-pro-48gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-m5-pro-48gb", "model": "/api/v1/models/gemma-4-31b-v2", "recipe": "/api/v1/recipes/gemma-4-31b-v2-int4-apple-m5-pro-48gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-31b-v2-int4-apple-m5-pro-48gb-mlx-tp1-sweep" }
2026-07-03
{ "best_ttft_ms": 596, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 33.4, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 250.7, "peak_vram_gb": 34.2, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-31b-v2", "name": "gemma-4-31B-v2" }
{ "format": "MLX", "id": "jdfelo-gemma-4-31b-v2-mlx-4bit--int4", "precision": "int4", "repository": "jdfelo/gemma-4-31B-v2-MLX-4bit" }
gemma-4-31b-v2-int4-apple-m5-pro-48gb-mlx-tp1
gemma-4-31b-v2-int4-apple-m5-pro-48gb-mlx-tp1-sweep
null
{ "name": "vllm", "version": null }
{ "count": 1, "id": "rx-7700-xt-12gb", "memory_gb": 12, "name": "AMD Radeon RX 7700 XT", "total_memory_gb": 12, "vendor": "amd" }
gemma-4-e4b-q4-k-m-rx-7700-xt-12gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rx-7700-xt-12gb", "model": "/api/v1/models/gemma-4-e4b", "recipe": "/api/v1/recipes/gemma-4-e4b-q4-k-m-rx-7700-xt-12gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/gemma-4-e4b-q4-k-m-rx-7700-xt-12gb-vllm-tp1-sweep" }
2026-07-25
{ "best_ttft_ms": 117.99, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 71.2, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 313.6, "peak_vram_gb": null, "point_count": 1 }
{ "family": "gemma", "id": "gemma-4-e4b", "name": "gemma-4-E4B" }
{ "format": "safetensors", "id": "google-gemma-4-e4b--q4-k-m", "precision": "Q4_K_M", "repository": "google/gemma-4-E4B" }
gemma-4-e4b-q4-k-m-rx-7700-xt-12gb-vllm-tp1
gemma-4-e4b-q4-k-m-rx-7700-xt-12gb-vllm-tp1-sweep
null
{ "name": "vllm", "version": "0.20.2+rocm723" }
{ "count": 3, "id": "radeon-ai-pro-r9700-32gb", "memory_gb": 32, "name": "Radeon AI PRO R9700", "total_memory_gb": 96, "vendor": "amd" }
glm-4-5-air-awq-4bit-radeon-ai-pro-r9700-32gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/radeon-ai-pro-r9700-32gb", "model": "/api/v1/models/glm-4-5-air", "recipe": "/api/v1/recipes/glm-4-5-air-awq-4bit-radeon-ai-pro-r9700-32gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/glm-4-5-air-awq-4bit-radeon-ai-pro-r9700-32gb-vllm-tp1-sweep" }
2026-07-02
{ "best_ttft_ms": 1958.56, "max_concurrency": 128, "max_context_tokens": 16384, "peak_decode_tok_s": 540.872, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "glm", "id": "glm-4-5-air", "name": "GLM-4.5-Air" }
{ "format": "safetensors", "id": "cyankiwi-glm-4-5-air-awq-4bit--awq-4bit", "precision": "AWQ-4bit", "repository": "cyankiwi/GLM-4.5-Air-AWQ-4bit" }
glm-4-5-air-awq-4bit-radeon-ai-pro-r9700-32gb-vllm-tp1
glm-4-5-air-awq-4bit-radeon-ai-pro-r9700-32gb-vllm-tp1-sweep
null
{ "name": "vllm", "version": "vllm-0.25.0" }
{ "count": 1, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 96, "vendor": "nvidia" }
glm-4-5-air-awq-4bit-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/glm-4-5-air", "recipe": "/api/v1/recipes/glm-4-5-air-awq-4bit-rtx-pro-6000-blackwell-96gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/glm-4-5-air-awq-4bit-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep" }
2026-07-14
{ "best_ttft_ms": 38.65, "max_concurrency": 96, "max_context_tokens": 32768, "peak_decode_tok_s": 2527.68, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "glm", "id": "glm-4-5-air", "name": "GLM-4.5-Air" }
{ "format": "safetensors", "id": "zai-org-glm-4-5-air--awq-4bit", "precision": "AWQ-4bit", "repository": "zai-org/GLM-4.5-Air" }
glm-4-5-air-awq-4bit-rtx-pro-6000-blackwell-96gb-vllm-tp1
glm-4-5-air-awq-4bit-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
null
{ "name": "mlx", "version": "0.31.3" }
{ "count": 1, "id": "apple-m5-pro-64gb", "memory_gb": 64, "name": "Apple M5 Pro 64GB", "total_memory_gb": 64, "vendor": "apple" }
glm-4-7-flash-4bit-apple-m5-pro-64gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-m5-pro-64gb", "model": "/api/v1/models/glm-4-7-flash", "recipe": "/api/v1/recipes/glm-4-7-flash-4bit-apple-m5-pro-64gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/glm-4-7-flash-4bit-apple-m5-pro-64gb-mlx-tp1-sweep" }
2026-04-29
{ "best_ttft_ms": 265.6, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 77.528, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": 17.208, "point_count": 1 }
{ "family": "glm", "id": "glm-4-7-flash", "name": "GLM-4.7-Flash" }
{ "format": "MLX", "id": "mlx-community-glm-4-7-flash-4bit--4bit", "precision": "4bit", "repository": "mlx-community/GLM-4.7-Flash-4bit" }
glm-4-7-flash-4bit-apple-m5-pro-64gb-mlx-tp1
glm-4-7-flash-4bit-apple-m5-pro-64gb-mlx-tp1-sweep
null
{ "name": "vllm", "version": "0.20.2+rocm723" }
{ "count": 3, "id": "radeon-ai-pro-r9700-32gb", "memory_gb": 32, "name": "Radeon AI PRO R9700", "total_memory_gb": 96, "vendor": "amd" }
glm-4-7-flash-gptq-int4-int8mix-radeon-ai-pro-r9700-32gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/radeon-ai-pro-r9700-32gb", "model": "/api/v1/models/glm-4-7-flash", "recipe": "/api/v1/recipes/glm-4-7-flash-gptq-int4-int8mix-radeon-ai-pro-r9700-32gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/glm-4-7-flash-gptq-int4-int8mix-radeon-ai-pro-r9700-32gb-vllm-tp1-sweep" }
2026-05-16
{ "best_ttft_ms": 304.27, "max_concurrency": 16, "max_context_tokens": 32768, "peak_decode_tok_s": 212.202, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "glm", "id": "glm-4-7-flash", "name": "GLM-4.7-Flash" }
{ "format": "safetensors", "id": "zai-org-glm-4-7-flash--gptq-int4-int8mix", "precision": "GPTQ-Int4-Int8Mix", "repository": "zai-org/GLM-4.7-Flash" }
glm-4-7-flash-gptq-int4-int8mix-radeon-ai-pro-r9700-32gb-vllm-tp1
glm-4-7-flash-gptq-int4-int8mix-radeon-ai-pro-r9700-32gb-vllm-tp1-sweep
null
{ "name": "vllm", "version": null }
{ "count": 1, "id": "rtx-5090-32gb", "memory_gb": 32, "name": "GeForce RTX 5090", "total_memory_gb": 32, "vendor": "nvidia" }
glm-4-7-flash-int4-rtx-5090-32gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-5090-32gb", "model": "/api/v1/models/glm-4-7-flash", "recipe": "/api/v1/recipes/glm-4-7-flash-int4-rtx-5090-32gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/glm-4-7-flash-int4-rtx-5090-32gb-vllm-tp1-sweep" }
2026-07-23
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 163.4, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "glm", "id": "glm-4-7-flash", "name": "GLM-4.7-Flash" }
{ "format": "safetensors", "id": "quanttrio-glm-4-7-flash-awq--int4", "precision": "INT4", "repository": "QuantTrio/GLM-4.7-Flash-AWQ" }
glm-4-7-flash-int4-rtx-5090-32gb-vllm-tp1
glm-4-7-flash-int4-rtx-5090-32gb-vllm-tp1-sweep
null
{ "name": "vllm", "version": null }
{ "count": 1, "id": "rx-7700-xt-12gb", "memory_gb": 12, "name": "AMD Radeon RX 7700 XT", "total_memory_gb": 12, "vendor": "amd" }
glm-4-7-flash-q4-k-m-rx-7700-xt-12gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rx-7700-xt-12gb", "model": "/api/v1/models/glm-4-7-flash", "recipe": "/api/v1/recipes/glm-4-7-flash-q4-k-m-rx-7700-xt-12gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/glm-4-7-flash-q4-k-m-rx-7700-xt-12gb-vllm-tp1-sweep" }
2026-07-16
{ "best_ttft_ms": 102.11, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 21.5, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 264.4, "peak_vram_gb": null, "point_count": 1 }
{ "family": "glm", "id": "glm-4-7-flash", "name": "GLM-4.7-Flash" }
{ "format": "safetensors", "id": "zai-org-glm-4-7-flash--q4-k-m", "precision": "Q4_K_M", "repository": "zai-org/GLM-4.7-Flash" }
glm-4-7-flash-q4-k-m-rx-7700-xt-12gb-vllm-tp1
glm-4-7-flash-q4-k-m-rx-7700-xt-12gb-vllm-tp1-sweep
null
{ "name": "sglang", "version": "0.5.10.post2.dev584+gc314a13f7" }
{ "count": 8, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 768, "vendor": "nvidia" }
glm-5-1-nvfp4-nvfp4-rtx-pro-6000-blackwell-96gb-sglang-tp8-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/glm-5-1-nvfp4", "recipe": "/api/v1/recipes/glm-5-1-nvfp4-nvfp4-rtx-pro-6000-blackwell-96gb-sglang-tp8", "speed_sweep": "/api/v1/speed-sweeps/glm-5-1-nvfp4-nvfp4-rtx-pro-6000-blackwell-96gb-sglang-tp8-sweep" }
2026-04-28
{ "best_ttft_ms": 20956, "max_concurrency": 10, "max_context_tokens": 200000, "peak_decode_tok_s": 74.53, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": 691, "point_count": 1 }
{ "family": "glm", "id": "glm-5-1-nvfp4", "name": "GLM-5.1-NVFP4" }
{ "format": "safetensors", "id": "lukealonso-glm-5-1-nvfp4-mtp--nvfp4", "precision": "NVFP4", "repository": "lukealonso/GLM-5.1-NVFP4-MTP" }
glm-5-1-nvfp4-nvfp4-rtx-pro-6000-blackwell-96gb-sglang-tp8
glm-5-1-nvfp4-nvfp4-rtx-pro-6000-blackwell-96gb-sglang-tp8-sweep
null
{ "name": "mlx", "version": "avlp12/mlx-lm glm-5.2-dsa-inference (f48df86)" }
{ "count": 1, "id": "apple-m3-ultra-512gb", "memory_gb": 512, "name": "Apple M3 Ultra 512GB", "total_memory_gb": 512, "vendor": "apple" }
glm-5-2-2-56bpw-mixed-apple-m3-ultra-512gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-m3-ultra-512gb", "model": "/api/v1/models/glm-5-2", "recipe": "/api/v1/recipes/glm-5-2-2-56bpw-mixed-apple-m3-ultra-512gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/glm-5-2-2-56bpw-mixed-apple-m3-ultra-512gb-mlx-tp1-sweep" }
2026-07-06
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 4096, "peak_decode_tok_s": 20.48, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 150.8, "peak_vram_gb": 245.5, "point_count": 1 }
{ "family": "glm", "id": "glm-5-2", "name": "GLM-5.2" }
{ "format": "MLX", "id": "avlp12-glm-5-2-alis-mlx-dynamic-2-56bpw--2-56bpw-mixed", "precision": "2.56bpw-mixed", "repository": "avlp12/GLM-5.2-Alis-MLX-Dynamic-2.56bpw" }
glm-5-2-2-56bpw-mixed-apple-m3-ultra-512gb-mlx-tp1
glm-5-2-2-56bpw-mixed-apple-m3-ultra-512gb-mlx-tp1-sweep
null
{ "name": "mlx", "version": "avlp12/mlx-lm glm-5.2-dsa-inference (f48df86)" }
{ "count": 1, "id": "apple-m3-ultra-512gb", "memory_gb": 512, "name": "Apple M3 Ultra 512GB", "total_memory_gb": 512, "vendor": "apple" }
glm-5-2-3-5bpw-mixed-apple-m3-ultra-512gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-m3-ultra-512gb", "model": "/api/v1/models/glm-5-2", "recipe": "/api/v1/recipes/glm-5-2-3-5bpw-mixed-apple-m3-ultra-512gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/glm-5-2-3-5bpw-mixed-apple-m3-ultra-512gb-mlx-tp1-sweep" }
2026-07-06
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 4096, "peak_decode_tok_s": 18.48, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 102.1, "peak_vram_gb": 336.1, "point_count": 1 }
{ "family": "glm", "id": "glm-5-2", "name": "GLM-5.2" }
{ "format": "MLX", "id": "avlp12-glm-5-2-alis-mlx-dynamic-3-5bpw--3-5bpw-mixed", "precision": "3.5bpw-mixed", "repository": "avlp12/GLM-5.2-Alis-MLX-Dynamic-3.5bpw" }
glm-5-2-3-5bpw-mixed-apple-m3-ultra-512gb-mlx-tp1
glm-5-2-3-5bpw-mixed-apple-m3-ultra-512gb-mlx-tp1-sweep
null
{ "name": "mlx", "version": null }
{ "count": 1, "id": "apple-m3-ultra-512gb", "memory_gb": 512, "name": "Apple M3 Ultra 512GB", "total_memory_gb": 512, "vendor": "apple" }
glm-5-2-mixed-3-6bit-apple-m3-ultra-512gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-m3-ultra-512gb", "model": "/api/v1/models/glm-5-2", "recipe": "/api/v1/recipes/glm-5-2-mixed-3-6bit-apple-m3-ultra-512gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/glm-5-2-mixed-3-6bit-apple-m3-ultra-512gb-mlx-tp1-sweep" }
2026-06-19
{ "best_ttft_ms": 80062, "max_concurrency": 1, "max_context_tokens": 8192, "peak_decode_tok_s": 14.8, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "glm", "id": "glm-5-2", "name": "GLM-5.2" }
{ "format": "MLX", "id": "pipenetwork-glm-5-2-mlx-mixed-3-6bit--mixed-3-6bit", "precision": "mixed-3_6bit", "repository": "pipenetwork/GLM-5.2-MLX-mixed-3_6bit" }
glm-5-2-mixed-3-6bit-apple-m3-ultra-512gb-mlx-tp1
glm-5-2-mixed-3-6bit-apple-m3-ultra-512gb-mlx-tp1-sweep
null
{ "name": "mlx", "version": null }
{ "count": 1, "id": "apple-m3-ultra-512gb", "memory_gb": 512, "name": "Apple M3 Ultra 512GB", "total_memory_gb": 512, "vendor": "apple" }
glm-5-2-mlx-2-32bpw-dynamic-apple-m3-ultra-512gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-m3-ultra-512gb", "model": "/api/v1/models/glm-5-2", "recipe": "/api/v1/recipes/glm-5-2-mlx-2-32bpw-dynamic-apple-m3-ultra-512gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/glm-5-2-mlx-2-32bpw-dynamic-apple-m3-ultra-512gb-mlx-tp1-sweep" }
2026-07-12
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 19.802, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 163.676, "peak_vram_gb": 222.873, "point_count": 1 }
{ "family": "glm", "id": "glm-5-2", "name": "GLM-5.2" }
{ "format": "MLX", "id": "avlp12-glm-5-2-alis-mlx-dynamic-2-3bpw--mlx-2-32bpw-dynamic", "precision": "MLX-2.32bpw-dynamic", "repository": "avlp12/GLM-5.2-Alis-MLX-Dynamic-2.3bpw" }
glm-5-2-mlx-2-32bpw-dynamic-apple-m3-ultra-512gb-mlx-tp1
glm-5-2-mlx-2-32bpw-dynamic-apple-m3-ultra-512gb-mlx-tp1-sweep
null
{ "name": "mlx", "version": null }
{ "count": 1, "id": "apple-m3-ultra-512gb", "memory_gb": 512, "name": "Apple M3 Ultra 512GB", "total_memory_gb": 512, "vendor": "apple" }
glm-5-2-mlx-2-56bpw-dynamic-apple-m3-ultra-512gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-m3-ultra-512gb", "model": "/api/v1/models/glm-5-2", "recipe": "/api/v1/recipes/glm-5-2-mlx-2-56bpw-dynamic-apple-m3-ultra-512gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/glm-5-2-mlx-2-56bpw-dynamic-apple-m3-ultra-512gb-mlx-tp1-sweep" }
2026-07-14
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 19.721, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 161.287, "peak_vram_gb": 245.522, "point_count": 1 }
{ "family": "glm", "id": "glm-5-2", "name": "GLM-5.2" }
{ "format": "MLX", "id": "avlp12-glm-5-2-alis-mlx-dynamic-2-56bpw--mlx-2-56bpw-dynamic", "precision": "MLX-2.56bpw-dynamic", "repository": "avlp12/GLM-5.2-Alis-MLX-Dynamic-2.56bpw" }
glm-5-2-mlx-2-56bpw-dynamic-apple-m3-ultra-512gb-mlx-tp1
glm-5-2-mlx-2-56bpw-dynamic-apple-m3-ultra-512gb-mlx-tp1-sweep
null
{ "name": "vllm", "version": "0.11.2.dev279+eldritch.enlightenment.v56a5c3e.kingdom.plus5.71fca68.b12x7bfc945.fi25dd814.cu132.20260706" }
{ "count": 4, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 384, "vendor": "nvidia" }
glm-5-2-mxfp8-nvfp4-nf3-hybrid-mxfp8-nvfp4-nf3-hybrid-rtx-pro-6000-blackwell-96gb-vllm-tp4-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/glm-5-2-mxfp8-nvfp4-nf3-hybrid", "recipe": "/api/v1/recipes/glm-5-2-mxfp8-nvfp4-nf3-hybrid-mxfp8-nvfp4-nf3-hybrid-rtx-pro-6000-blackwell-96gb-vllm-tp4", "speed_sweep": "/api/v1/speed-sweeps/glm-5-2-mxfp8-nvfp4-nf3-hybrid-mxfp...
2026-07-31
{ "best_ttft_ms": 355.0544859899674, "max_concurrency": 1, "max_context_tokens": 480000, "peak_decode_tok_s": 91.96917531860531, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "glm", "id": "glm-5-2-mxfp8-nvfp4-nf3-hybrid", "name": "GLM-5.2-MXFP8-NVFP4-NF3-Hybrid" }
{ "format": "safetensors", "id": "jacklarmer-glm-5-2-abliterated-mxfp8-nvfp4-nf3-hybrid--mxfp8-nvfp4-nf3-hybrid", "precision": "MXFP8-NVFP4-NF3-Hybrid", "repository": "jacklarmer/GLM-5.2-Abliterated-MXFP8-NVFP4-NF3-Hybrid" }
glm-5-2-mxfp8-nvfp4-nf3-hybrid-mxfp8-nvfp4-nf3-hybrid-rtx-pro-6000-blackwell-96gb-vllm-tp4
glm-5-2-mxfp8-nvfp4-nf3-hybrid-mxfp8-nvfp4-nf3-hybrid-rtx-pro-6000-blackwell-96gb-vllm-tp4-sweep
null
{ "name": "vllm", "version": null }
{ "count": 4, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 384, "vendor": "nvidia" }
glm-5-2-nvfp4-nvfp4-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/glm-5-2-nvfp4", "recipe": "/api/v1/recipes/glm-5-2-nvfp4-nvfp4-rtx-pro-6000-blackwell-96gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/glm-5-2-nvfp4-nvfp4-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep" }
2026-06-29
{ "best_ttft_ms": 138.4989789999622, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 52.55232073588523, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 1850.956966361186, "peak_vram_gb": null, "point_count": 1 }
{ "family": "glm", "id": "glm-5-2-nvfp4", "name": "GLM-5.2-NVFP4" }
{ "format": "safetensors", "id": "0xsero-glm-5-2-504b-nvidia--nvfp4", "precision": "NVFP4", "repository": "0xSero/GLM-5.2-504B-Nvidia" }
glm-5-2-nvfp4-nvfp4-rtx-pro-6000-blackwell-96gb-vllm-tp1
glm-5-2-nvfp4-nvfp4-rtx-pro-6000-blackwell-96gb-vllm-tp1-sweep
null
{ "name": "vllm", "version": "0.11.2.dev279+dark.devotion.pr31.pr15.w4a16scale" }
{ "count": 4, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 384, "vendor": "nvidia" }
glm-5-2-nvfp4-nvfp4-rtx-pro-6000-blackwell-96gb-vllm-tp4-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/glm-5-2-nvfp4", "recipe": "/api/v1/recipes/glm-5-2-nvfp4-nvfp4-rtx-pro-6000-blackwell-96gb-vllm-tp4", "speed_sweep": "/api/v1/speed-sweeps/glm-5-2-nvfp4-nvfp4-rtx-pro-6000-blackwell-96gb-vllm-tp4-sweep" }
2026-07-25
{ "best_ttft_ms": 356.7, "max_concurrency": 1, "max_context_tokens": 1000000, "peak_decode_tok_s": 87.09, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 1555.91, "peak_vram_gb": 93.4, "point_count": 1 }
{ "family": "glm", "id": "glm-5-2-nvfp4", "name": "GLM-5.2-NVFP4" }
{ "format": "safetensors", "id": "madeby561-glm-5-2-nvfp4-reap-504b-term--nvfp4", "precision": "NVFP4", "repository": "madeby561/GLM-5.2-NVFP4-REAP-504B-term" }
glm-5-2-nvfp4-nvfp4-rtx-pro-6000-blackwell-96gb-vllm-tp4
glm-5-2-nvfp4-nvfp4-rtx-pro-6000-blackwell-96gb-vllm-tp4-sweep
null
{ "name": "vllm", "version": null }
{ "count": 4, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 384, "vendor": "nvidia" }
glm-5-3-flash-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp4-7mfqe264-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/glm-5-3-flash", "recipe": "/api/v1/recipes/glm-5-3-flash-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp4-7mfqe264", "speed_sweep": "/api/v1/speed-sweeps/glm-5-3-flash-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp4-7mfqe264-sweep" }
2026-08-26T20:15:45.070Z
{ "best_ttft_ms": 90.27875200081326, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 146.7948382373735, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 7059.962295978053, "peak_vram_gb": null, "point_count": 1 }
{ "family": "glm", "id": "glm-5-3-flash", "name": "GLM-5.3-Flash" }
{ "format": "FP8", "id": "zai-org-glm-5-3-flash--fp8", "precision": "FP8", "repository": "zai-org/GLM-5.3-Flash" }
glm-5-3-flash-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp4-7mfqe264
glm-5-3-flash-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp4-7mfqe264-sweep
2026-08-24T00:00:00
{ "name": "vllm", "version": "controller-managed; captured 2026-08-23" }
{ "count": 4, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 384, "vendor": "nvidia" }
glm-52-vision-exl3-tr3-3bpw-vision-rtxpro6000-vllm-tp4-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/glm-5-2-vision", "recipe": "/api/v1/recipes/glm-52-vision-exl3-tr3-3bpw-vision-rtxpro6000-vllm-tp4", "speed_sweep": "/api/v1/speed-sweeps/glm-52-vision-exl3-tr3-3bpw-vision-rtxpro6000-vllm-tp4-sweep" }
2026-08-23
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 400000, "peak_decode_tok_s": 70.698, "peak_decode_tok_s_per_stream": 70.698, "peak_prefill_tok_s": 1804.34, "peak_vram_gb": null, "point_count": 4 }
{ "family": "glm", "id": "glm-5-2-vision", "name": "GLM-5.2-Vision" }
{ "format": "EXL3 TR3 text + BF16 MoonViT vision graft", "id": "0xsero-glm-5-2-tr3-vision--3-0-bpw-text-bf16-vision", "precision": "3.0 bpw text + BF16 vision", "repository": "0xSero/GLM-5.2-TR3-Vision" }
glm-52-vision-exl3-tr3-3bpw-vision-rtxpro6000-vllm-tp4
glm-52-vision-exl3-tr3-3bpw-vision-rtxpro6000-vllm-tp4-sweep
2026-08-23T00:00:00
{ "name": "vllm", "version": "0.11.2.dev280+gilded.gnosis.v20.vllmb46c3aa.si35aebc6.fi801d57a.cu132.20260730.r12" }
{ "count": 4, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 384, "vendor": "nvidia" }
glm52-exl3-rtxpro6000-vllm-tp4-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/glm-5-2", "recipe": "/api/v1/recipes/glm52-exl3-rtxpro6000-vllm-tp4", "speed_sweep": "/api/v1/speed-sweeps/glm52-exl3-rtxpro6000-vllm-tp4-sweep" }
2026-08-17
{ "best_ttft_ms": null, "max_concurrency": 4, "max_context_tokens": 8192, "peak_decode_tok_s": 403.8167, "peak_decode_tok_s_per_stream": 188.1317, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 3 }
{ "family": "glm", "id": "glm-5-2", "name": "GLM-5.2" }
{ "format": "EXL3 Trellis TR3", "id": "brandonmusic-glm-5-2-exl3-tr3-3-0bpw--3-0-bpw", "precision": "3.0 bpw", "repository": "brandonmusic/GLM-5.2-EXL3-TR3-3.0bpw" }
glm52-exl3-rtxpro6000-vllm-tp4
glm52-exl3-rtxpro6000-vllm-tp4-sweep
2026-08-23T00:00:00
{ "name": "vllm", "version": "0.27.0 + b12x 334a2d75d166becea0aa640b402d521ea0a290eb" }
{ "count": 4, "id": "dgx-spark-gb10-128gb", "memory_gb": 128, "name": "NVIDIA DGX Spark GB10", "total_memory_gb": 512, "vendor": "nvidia" }
glm52-int4-int8mix-dgxspark-vllm-tp4-sweep
{ "hardware": "/api/v1/hardware/dgx-spark-gb10-128gb", "model": "/api/v1/models/glm-5-2", "recipe": "/api/v1/recipes/glm52-int4-int8mix-dgxspark-vllm-tp4", "speed_sweep": "/api/v1/speed-sweeps/glm52-int4-int8mix-dgxspark-vllm-tp4-sweep" }
2026-08-16
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": null, "peak_decode_tok_s": null, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 618, "peak_vram_gb": null, "point_count": 1 }
{ "family": "glm", "id": "glm-5-2", "name": "GLM-5.2" }
{ "format": "compressed-tensors", "id": "quanttrio-glm-5-2-int4-int8mix--w4a16-int8mix", "precision": "W4A16 + Int8Mix", "repository": "QuantTrio/GLM-5.2-Int4-Int8Mix" }
glm52-int4-int8mix-dgxspark-vllm-tp4
glm52-int4-int8mix-dgxspark-vllm-tp4-sweep
2026-08-24T00:00:00
{ "name": "vllm", "version": "c382f1d28d5be2f867c216609408bdb424d6049a+b12x-e44cb77777a075790ebe9f7aa9f225d073aea109" }
{ "count": 4, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 384, "vendor": "nvidia" }
glm52-mxfp8-nvfp4-nf3-rtxpro6000-vllm-tp4-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/glm-5-2", "recipe": "/api/v1/recipes/glm52-mxfp8-nvfp4-nf3-rtxpro6000-vllm-tp4", "speed_sweep": "/api/v1/speed-sweeps/glm52-mxfp8-nvfp4-nf3-rtxpro6000-vllm-tp4-sweep" }
2026-07-29
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 180007, "peak_decode_tok_s": 58.2, "peak_decode_tok_s_per_stream": 58.2, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "glm", "id": "glm-5-2", "name": "GLM-5.2" }
{ "format": "safetensors", "id": "madeby561-glm-5-2-mxfp8-nvfp4-nf3-hybrid--mxfp8-nvfp4-nf3-hybrid", "precision": "MXFP8/NVFP4/NF3 hybrid", "repository": "madeby561/GLM-5.2-MXFP8-NVFP4-NF3-Hybrid" }
glm52-mxfp8-nvfp4-nf3-rtxpro6000-vllm-tp4
glm52-mxfp8-nvfp4-nf3-rtxpro6000-vllm-tp4-sweep
2026-08-23T00:00:00
{ "name": "vllm", "version": "jarrelscy glm52-sm120 fork" }
{ "count": 3, "id": "dgx-spark-gb10-128gb", "memory_gb": 128, "name": "NVIDIA DGX Spark GB10", "total_memory_gb": 384, "vendor": "nvidia" }
glm52-nvfp4-aqlm-dgxspark-vllm-tp3-sweep
{ "hardware": "/api/v1/hardware/dgx-spark-gb10-128gb", "model": "/api/v1/models/glm-5-2", "recipe": "/api/v1/recipes/glm52-nvfp4-aqlm-dgxspark-vllm-tp3", "speed_sweep": "/api/v1/speed-sweeps/glm52-nvfp4-aqlm-dgxspark-vllm-tp3-sweep" }
2026-07-26
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 40000, "peak_decode_tok_s": 21, "peak_decode_tok_s_per_stream": 21, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 2 }
{ "family": "glm", "id": "glm-5-2", "name": "GLM-5.2" }
{ "format": "hybrid ModelOpt+AQLM", "id": "jarrelscy-glm-5-2-nvfp4-aqlm-hybrid--nvfp4-hot-experts-aqlm-2-bit-cold-experts", "precision": "NVFP4 hot experts + AQLM 2-bit cold experts", "repository": "jarrelscy/GLM-5.2-NVFP4-AQLM-hybrid" }
glm52-nvfp4-aqlm-dgxspark-vllm-tp3
glm52-nvfp4-aqlm-dgxspark-vllm-tp3-sweep
2026-08-24T00:00:00
{ "name": "vllm", "version": "voipmonitor b12x build 20260608; exact commit unreported" }
{ "count": 4, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 384, "vendor": "nvidia" }
glm52-nvfp4-reap-469b-rtxpro6000-vllm-tp4-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/glm-5-2", "recipe": "/api/v1/recipes/glm52-nvfp4-reap-469b-rtxpro6000-vllm-tp4", "speed_sweep": "/api/v1/speed-sweeps/glm52-nvfp4-reap-469b-rtxpro6000-vllm-tp4-sweep" }
2026-08-24
{ "best_ttft_ms": 12000, "max_concurrency": 1, "max_context_tokens": 64000, "peak_decode_tok_s": 60, "peak_decode_tok_s_per_stream": 60, "peak_prefill_tok_s": 5100, "peak_vram_gb": null, "point_count": 2 }
{ "family": "glm", "id": "glm-5-2", "name": "GLM-5.2" }
{ "format": "ModelOpt safetensors", "id": "0xsero-glm-5-2-nvfp4-reap-469b--nvfp4-reap", "precision": "NVFP4 REAP", "repository": "0xSero/GLM-5.2-NVFP4-REAP-469B" }
glm52-nvfp4-reap-469b-rtxpro6000-vllm-tp4
glm52-nvfp4-reap-469b-rtxpro6000-vllm-tp4-sweep
2026-08-24T00:00:00
{ "name": "vllm", "version": "b12x nonuniform fork 20260705; exact commit unreported" }
{ "count": 4, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 384, "vendor": "nvidia" }
glm52-nvfp4-reap-nu176-rtxpro6000-vllm-tp4-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/glm-5-2", "recipe": "/api/v1/recipes/glm52-nvfp4-reap-nu176-rtxpro6000-vllm-tp4", "speed_sweep": "/api/v1/speed-sweeps/glm52-nvfp4-reap-nu176-rtxpro6000-vllm-tp4-sweep" }
2026-08-24
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": null, "peak_decode_tok_s": 77, "peak_decode_tok_s_per_stream": 77, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "glm", "id": "glm-5-2", "name": "GLM-5.2" }
{ "format": "ModelOpt safetensors", "id": "0xsero-glm-5-2-reap-nu176-526b--nvfp4-non-uniform-reap", "precision": "NVFP4 non-uniform REAP", "repository": "0xSero/GLM-5.2-REAP-NU176-526B" }
glm52-nvfp4-reap-nu176-rtxpro6000-vllm-tp4
glm52-nvfp4-reap-nu176-rtxpro6000-vllm-tp4-sweep
2026-08-27T00:00:00
{ "name": "sglang", "version": "ghcr.io/0xsero/glm-5.3-flash-sglang-sm121@sha256:f9ac60ba4071f8acd64f0f3c074aca308f6d659405fee46fc8031489a1e8b19b" }
{ "count": 2, "id": "dgx-spark-gb10-128gb", "memory_gb": 128, "name": "NVIDIA DGX Spark GB10", "total_memory_gb": 256, "vendor": "nvidia" }
glm53-flash-nvfp4-dgxspark-sglang-tp2-sweep
{ "hardware": "/api/v1/hardware/dgx-spark-gb10-128gb", "model": "/api/v1/models/glm-5-3-flash", "recipe": "/api/v1/recipes/glm53-flash-nvfp4-dgxspark-sglang-tp2", "speed_sweep": "/api/v1/speed-sweeps/glm53-flash-nvfp4-dgxspark-sglang-tp2-sweep" }
2026-08-27
{ "best_ttft_ms": null, "max_concurrency": 2, "max_context_tokens": 131072, "peak_decode_tok_s": 30.72, "peak_decode_tok_s_per_stream": 17.49, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 2 }
{ "family": "glm", "id": "glm-5-3-flash", "name": "GLM-5.3-Flash" }
{ "format": "ModelOpt", "id": "libertaidai-glm-5-3-flash-nvfp4--nvfp4", "precision": "NVFP4", "repository": "LibertAIDAI/GLM-5.3-Flash-NVFP4" }
glm53-flash-nvfp4-dgxspark-sglang-tp2
glm53-flash-nvfp4-dgxspark-sglang-tp2-sweep
2026-08-27T00:00:00
{ "name": "sglang", "version": "ghcr.io/0xsero/glm-5.3-flash-sglang-sm121@sha256:f9ac60ba4071f8acd64f0f3c074aca308f6d659405fee46fc8031489a1e8b19b" }
{ "count": 4, "id": "dgx-spark-gb10-128gb", "memory_gb": 128, "name": "NVIDIA DGX Spark GB10", "total_memory_gb": 512, "vendor": "nvidia" }
glm53-flash-nvfp4-dgxspark-sglang-tp4-sweep
{ "hardware": "/api/v1/hardware/dgx-spark-gb10-128gb", "model": "/api/v1/models/glm-5-3-flash", "recipe": "/api/v1/recipes/glm53-flash-nvfp4-dgxspark-sglang-tp4", "speed_sweep": "/api/v1/speed-sweeps/glm53-flash-nvfp4-dgxspark-sglang-tp4-sweep" }
2026-08-27
{ "best_ttft_ms": null, "max_concurrency": 8, "max_context_tokens": 1048576, "peak_decode_tok_s": 82.37, "peak_decode_tok_s_per_stream": 22.69, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 3 }
{ "family": "glm", "id": "glm-5-3-flash", "name": "GLM-5.3-Flash" }
{ "format": "ModelOpt", "id": "libertaidai-glm-5-3-flash-nvfp4--nvfp4", "precision": "NVFP4", "repository": "LibertAIDAI/GLM-5.3-Flash-NVFP4" }
glm53-flash-nvfp4-dgxspark-sglang-tp4
glm53-flash-nvfp4-dgxspark-sglang-tp4-sweep
null
{ "name": "sglang", "version": "lmsysorg/sglang:glm-5.3-flash@sha256:3a97bd50034ca60c6e6c86b8e36a73675d261f6a5eb71197796aee5175409290" }
{ "count": 4, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 384, "vendor": "nvidia" }
glm53-flash-nvfp4-rtxpro6000-sglang-tp4-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/glm-5-3-flash", "recipe": "/api/v1/recipes/glm53-flash-nvfp4-rtxpro6000-sglang-tp4", "speed_sweep": "/api/v1/speed-sweeps/glm53-flash-nvfp4-rtxpro6000-sglang-tp4-sweep" }
2026-08-26
{ "best_ttft_ms": 124, "max_concurrency": 8, "max_context_tokens": 213094, "peak_decode_tok_s": 550, "peak_decode_tok_s_per_stream": 141.8, "peak_prefill_tok_s": 6383, "peak_vram_gb": null, "point_count": 10 }
{ "family": "glm", "id": "glm-5-3-flash", "name": "GLM-5.3-Flash" }
{ "format": "ModelOpt", "id": "libertaidai-glm-5-3-flash-nvfp4--nvfp4", "precision": "NVFP4", "repository": "LibertAIDAI/GLM-5.3-Flash-NVFP4" }
glm53-flash-nvfp4-rtxpro6000-sglang-tp4
glm53-flash-nvfp4-rtxpro6000-sglang-tp4-sweep
null
{ "name": "vllm", "version": "0.19.2rc1.dev134+gfe9c3d6c5" }
{ "count": 1, "id": "rtx-5090-32gb", "memory_gb": 32, "name": "GeForce RTX 5090", "total_memory_gb": 32, "vendor": "nvidia" }
huihui-qwen3-6-27b-abliterated-nvfp4-rtx-5090-32gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-5090-32gb", "model": "/api/v1/models/huihui-qwen3-6-27b-abliterated", "recipe": "/api/v1/recipes/huihui-qwen3-6-27b-abliterated-nvfp4-rtx-5090-32gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/huihui-qwen3-6-27b-abliterated-nvfp4-rtx-5090-32gb-vllm-tp1-sweep" }
2026-05-06
{ "best_ttft_ms": 766, "max_concurrency": 1, "max_context_tokens": 8379, "peak_decode_tok_s": 105.3, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 10940, "peak_vram_gb": 32, "point_count": 1 }
{ "family": "qwen", "id": "huihui-qwen3-6-27b-abliterated", "name": "Huihui-Qwen3.6-27B-abliterated" }
{ "format": "safetensors", "id": "sakamakismile-huihui-qwen3-6-27b-abliterated-nvfp4-mtp--nvfp4", "precision": "NVFP4", "repository": "sakamakismile/Huihui-Qwen3.6-27B-abliterated-NVFP4-MTP" }
huihui-qwen3-6-27b-abliterated-nvfp4-rtx-5090-32gb-vllm-tp1
huihui-qwen3-6-27b-abliterated-nvfp4-rtx-5090-32gb-vllm-tp1-sweep
null
{ "name": "mlx", "version": "oMLX 0.6.1 (12249931772fe070fb957a85a65c8f87ea4363b8)" }
{ "count": 1, "id": "apple-m3-ultra-96gb", "memory_gb": 96, "name": "Apple M3 Ultra 96GB", "total_memory_gb": 96, "vendor": "apple" }
huihui-qwen3-8-27b-abliterated-mlx-8bit-apple-m3-ultra-96gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-m3-ultra-96gb", "model": "/api/v1/models/huihui-qwen3-8-27b-abliterated", "recipe": "/api/v1/recipes/huihui-qwen3-8-27b-abliterated-mlx-8bit-apple-m3-ultra-96gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/huihui-qwen3-8-27b-abliterated-mlx-8bit-apple-m3-ultra-96gb-mlx-tp...
2026-08-19
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 262144, "peak_decode_tok_s": 38.3, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "qwen", "id": "huihui-qwen3-8-27b-abliterated", "name": "Huihui-Qwen3.8-27B-abliterated" }
{ "format": "MLX", "id": "ailexleon-huihui-qwen3-8-27b-abliterated-mlx-8bit--mlx-8bit", "precision": "MLX-8bit", "repository": "ailexleon/Huihui-Qwen3.8-27B-abliterated-mlx-8Bit" }
huihui-qwen3-8-27b-abliterated-mlx-8bit-apple-m3-ultra-96gb-mlx-tp1
huihui-qwen3-8-27b-abliterated-mlx-8bit-apple-m3-ultra-96gb-mlx-tp1-sweep
null
{ "name": "vllm", "version": null }
{ "count": 1, "id": "rtx-5090-32gb", "memory_gb": 32, "name": "GeForce RTX 5090", "total_memory_gb": 32, "vendor": "nvidia" }
huihui-qwen3-8-27b-abliterated-nvfp4-rtx-5090-32gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-5090-32gb", "model": "/api/v1/models/huihui-qwen3-8-27b-abliterated", "recipe": "/api/v1/recipes/huihui-qwen3-8-27b-abliterated-nvfp4-rtx-5090-32gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/huihui-qwen3-8-27b-abliterated-nvfp4-rtx-5090-32gb-vllm-tp1-sweep" }
2026-08-17
{ "best_ttft_ms": 186, "max_concurrency": 1, "max_context_tokens": 345, "peak_decode_tok_s": 178, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 1860, "peak_vram_gb": null, "point_count": 1 }
{ "family": "qwen", "id": "huihui-qwen3-8-27b-abliterated", "name": "Huihui-Qwen3.8-27B-abliterated" }
{ "format": "safetensors", "id": "sakamakismile-huihui-qwen3-8-27b-abliterated-nvfp4--nvfp4", "precision": "NVFP4", "repository": "sakamakismile/Huihui-Qwen3.8-27B-abliterated-NVFP4" }
huihui-qwen3-8-27b-abliterated-nvfp4-rtx-5090-32gb-vllm-tp1
huihui-qwen3-8-27b-abliterated-nvfp4-rtx-5090-32gb-vllm-tp1-sweep
2026-08-23T00:00:00
{ "name": "vllm", "version": "65b7662d3fcb773afaf751ab29ac6960a0cf011d+sm121-overlays" }
{ "count": 2, "id": "dgx-spark-gb10-128gb", "memory_gb": 128, "name": "NVIDIA DGX Spark GB10", "total_memory_gb": 256, "vendor": "nvidia" }
inkling-small-nvfp4-dgxspark-vllm-tp2-sweep
{ "hardware": "/api/v1/hardware/dgx-spark-gb10-128gb", "model": "/api/v1/models/inkling-small", "recipe": "/api/v1/recipes/inkling-small-nvfp4-dgxspark-vllm-tp2", "speed_sweep": "/api/v1/speed-sweeps/inkling-small-nvfp4-dgxspark-vllm-tp2-sweep" }
2026-08-12
{ "best_ttft_ms": 334.33, "max_concurrency": 4, "max_context_tokens": 16437, "peak_decode_tok_s": 81.4, "peak_decode_tok_s_per_stream": 32.31, "peak_prefill_tok_s": 3504.03, "peak_vram_gb": null, "point_count": 6 }
{ "family": "inkling", "id": "inkling-small", "name": "Inkling-Small" }
{ "format": "ModelOpt", "id": "thinkingmachines-inkling-small-nvfp4--nvfp4", "precision": "NVFP4", "repository": "thinkingmachines/Inkling-Small-NVFP4" }
inkling-small-nvfp4-dgxspark-vllm-tp2
inkling-small-nvfp4-dgxspark-vllm-tp2-sweep
null
{ "name": "sglang", "version": "lmsysorg/sglang:dev-cu13-inkling-dspark@sha256:fbea1a4e25b26660dbc2384a27ead8817e9b7670f257b5c3143e0450d14524d7 + 1 patched file (SGLang commit b7252cc6b0c78b25ecea7ee5efa91a6ae37d0f19); built image local/sglang-inkling:sm120" }
{ "count": 2, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 192, "vendor": "nvidia" }
inkling-small-nvfp4-rtxpro6000-sglang-tp2-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/inkling-small", "recipe": "/api/v1/recipes/inkling-small-nvfp4-rtxpro6000-sglang-tp2", "speed_sweep": "/api/v1/speed-sweeps/inkling-small-nvfp4-rtxpro6000-sglang-tp2-sweep" }
2026-08-26
{ "best_ttft_ms": 60, "max_concurrency": 4, "max_context_tokens": 50, "peak_decode_tok_s": 306.6, "peak_decode_tok_s_per_stream": 117.8, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 5 }
{ "family": "inkling", "id": "inkling-small", "name": "Inkling-Small" }
{ "format": "ModelOpt", "id": "thinkingmachines-inkling-small-nvfp4--nvfp4", "precision": "NVFP4", "repository": "thinkingmachines/Inkling-Small-NVFP4" }
inkling-small-nvfp4-rtxpro6000-sglang-tp2
inkling-small-nvfp4-rtxpro6000-sglang-tp2-sweep
null
{ "name": "SGLang", "version": "0.5.9" }
{ "count": 8, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 768, "vendor": "nvidia" }
kimi-k2-5-int4-rtx-pro-6000-blackwell-96gb-sglang-tp8-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/kimi-k2-5", "recipe": "/api/v1/recipes/kimi-k2-5-int4-rtx-pro-6000-blackwell-96gb-sglang-tp8", "speed_sweep": "/api/v1/speed-sweeps/kimi-k2-5-int4-rtx-pro-6000-blackwell-96gb-sglang-tp8-sweep" }
2026-04-25
{ "best_ttft_ms": 3710, "max_concurrency": 10, "max_context_tokens": 200000, "peak_decode_tok_s": 74, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": 748.88, "point_count": 1 }
{ "family": "kimi", "id": "kimi-k2-5", "name": "Kimi-K2.5" }
{ "format": "safetensors", "id": "moonshotai-kimi-k2-5--int4", "precision": "INT4", "repository": "moonshotai/Kimi-K2.5" }
kimi-k2-5-int4-rtx-pro-6000-blackwell-96gb-sglang-tp8
kimi-k2-5-int4-rtx-pro-6000-blackwell-96gb-sglang-tp8-sweep
null
{ "name": "mlx", "version": null }
{ "count": 1, "id": "apple-m5-pro-48gb", "memory_gb": 48, "name": "Apple M5 Pro 48GB", "total_memory_gb": 48, "vendor": "apple" }
lfm2-5-8b-a1b-base-int4-apple-m5-pro-48gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-m5-pro-48gb", "model": "/api/v1/models/lfm2-5-8b-a1b-base", "recipe": "/api/v1/recipes/lfm2-5-8b-a1b-base-int4-apple-m5-pro-48gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/lfm2-5-8b-a1b-base-int4-apple-m5-pro-48gb-mlx-tp1-sweep" }
2026-05-30
{ "best_ttft_ms": 133.6, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 189.3, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 179.6, "peak_vram_gb": 4.82, "point_count": 1 }
{ "family": "lfm", "id": "lfm2-5-8b-a1b-base", "name": "LFM2.5-8B-A1B-Base" }
{ "format": "MLX", "id": "liquidai-lfm2-5-8b-a1b--int4", "precision": "INT4", "repository": "LiquidAI/LFM2.5-8B-A1B" }
lfm2-5-8b-a1b-base-int4-apple-m5-pro-48gb-mlx-tp1
lfm2-5-8b-a1b-base-int4-apple-m5-pro-48gb-mlx-tp1-sweep
null
{ "name": "vllm", "version": null }
{ "count": 1, "id": "rx-7700-xt-12gb", "memory_gb": 12, "name": "AMD Radeon RX 7700 XT", "total_memory_gb": 12, "vendor": "amd" }
lfm2-5-8b-a1b-iq4-xs-rx-7700-xt-12gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/rx-7700-xt-12gb", "model": "/api/v1/models/lfm2-5-8b-a1b", "recipe": "/api/v1/recipes/lfm2-5-8b-a1b-iq4-xs-rx-7700-xt-12gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/lfm2-5-8b-a1b-iq4-xs-rx-7700-xt-12gb-vllm-tp1-sweep" }
2026-07-16
{ "best_ttft_ms": 64.8, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 156.7, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 478.4, "peak_vram_gb": null, "point_count": 1 }
{ "family": "lfm", "id": "lfm2-5-8b-a1b", "name": "LFM2.5-8B-A1B" }
{ "format": "safetensors", "id": "unsloth-lfm2-5-8b-a1b--iq4-xs", "precision": "IQ4_XS", "repository": "unsloth/LFM2.5-8B-A1B" }
lfm2-5-8b-a1b-iq4-xs-rx-7700-xt-12gb-vllm-tp1
lfm2-5-8b-a1b-iq4-xs-rx-7700-xt-12gb-vllm-tp1-sweep
2026-08-24T00:00:00
{ "name": "sglang", "version": "0.5.13.post1.dev20260623+g0460f277b7" }
{ "count": 1, "id": "mi300x-192gb", "memory_gb": 192, "name": "AMD Instinct MI300X", "total_memory_gb": 192, "vendor": "amd" }
lfm25-26b-bf16-mi300x-sglang-tp1-sweep
{ "hardware": "/api/v1/hardware/mi300x-192gb", "model": "/api/v1/models/lfm2-5-2-6b", "recipe": "/api/v1/recipes/lfm25-26b-bf16-mi300x-sglang-tp1", "speed_sweep": "/api/v1/speed-sweeps/lfm25-26b-bf16-mi300x-sglang-tp1-sweep" }
2026-08-24
{ "best_ttft_ms": 926.8, "max_concurrency": 4, "max_context_tokens": 130560, "peak_decode_tok_s": null, "peak_decode_tok_s_per_stream": 122.2, "peak_prefill_tok_s": 140747.8, "peak_vram_gb": null, "point_count": 6 }
{ "family": "lfm", "id": "lfm2-5-2-6b", "name": "LFM2.5-2.6B" }
{ "format": "safetensors", "id": "liquidai-lfm2-5-2-6b--bf16", "precision": "BF16", "repository": "LiquidAI/LFM2.5-2.6B" }
lfm25-26b-bf16-mi300x-sglang-tp1
lfm25-26b-bf16-mi300x-sglang-tp1-sweep
2026-08-24T00:00:00
{ "name": "sglang", "version": "0.0.0.dev1+geec794bce.d20260823" }
{ "count": 1, "id": "rtx-2000-ada-16gb", "memory_gb": 16, "name": "NVIDIA RTX 2000 Ada Generation", "total_memory_gb": 16, "vendor": "nvidia" }
lfm25-26b-bf16-rtx2000ada-sglang-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-2000-ada-16gb", "model": "/api/v1/models/lfm2-5-2-6b", "recipe": "/api/v1/recipes/lfm25-26b-bf16-rtx2000ada-sglang-tp1", "speed_sweep": "/api/v1/speed-sweeps/lfm25-26b-bf16-rtx2000ada-sglang-tp1-sweep" }
2026-08-24
{ "best_ttft_ms": 693.4, "max_concurrency": 2, "max_context_tokens": 130560, "peak_decode_tok_s": null, "peak_decode_tok_s_per_stream": 26.5, "peak_prefill_tok_s": 184386.1, "peak_vram_gb": null, "point_count": 4 }
{ "family": "lfm", "id": "lfm2-5-2-6b", "name": "LFM2.5-2.6B" }
{ "format": "safetensors", "id": "liquidai-lfm2-5-2-6b--bf16", "precision": "BF16", "repository": "LiquidAI/LFM2.5-2.6B" }
lfm25-26b-bf16-rtx2000ada-sglang-tp1
lfm25-26b-bf16-rtx2000ada-sglang-tp1-sweep
2026-08-24T00:00:00
{ "name": "sglang", "version": "0.0.0.dev1+geec794bce.d20260823" }
{ "count": 1, "id": "rtx-3090-24gb", "memory_gb": 24, "name": "GeForce RTX 3090", "total_memory_gb": 24, "vendor": "nvidia" }
lfm25-26b-bf16-rtx3090-sglang-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-3090-24gb", "model": "/api/v1/models/lfm2-5-2-6b", "recipe": "/api/v1/recipes/lfm25-26b-bf16-rtx3090-sglang-tp1", "speed_sweep": "/api/v1/speed-sweeps/lfm25-26b-bf16-rtx3090-sglang-tp1-sweep" }
2026-08-24
{ "best_ttft_ms": 394.1296, "max_concurrency": 4, "max_context_tokens": 130560, "peak_decode_tok_s": null, "peak_decode_tok_s_per_stream": 89.9464, "peak_prefill_tok_s": 328304.8029, "peak_vram_gb": null, "point_count": 6 }
{ "family": "lfm", "id": "lfm2-5-2-6b", "name": "LFM2.5-2.6B" }
{ "format": "safetensors", "id": "liquidai-lfm2-5-2-6b--bf16", "precision": "BF16", "repository": "LiquidAI/LFM2.5-2.6B" }
lfm25-26b-bf16-rtx3090-sglang-tp1
lfm25-26b-bf16-rtx3090-sglang-tp1-sweep
2026-08-24T00:00:00
{ "name": "sglang", "version": "0.0.0.dev1+geec794bce.d20260823" }
{ "count": 1, "id": "rtx-4000-ada-20gb", "memory_gb": 20, "name": "NVIDIA RTX 4000 Ada Generation", "total_memory_gb": 20, "vendor": "nvidia" }
lfm25-26b-bf16-rtx4000ada-sglang-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-4000-ada-20gb", "model": "/api/v1/models/lfm2-5-2-6b", "recipe": "/api/v1/recipes/lfm25-26b-bf16-rtx4000ada-sglang-tp1", "speed_sweep": "/api/v1/speed-sweeps/lfm25-26b-bf16-rtx4000ada-sglang-tp1-sweep" }
2026-08-24
{ "best_ttft_ms": 364.4, "max_concurrency": 2, "max_context_tokens": 130560, "peak_decode_tok_s": null, "peak_decode_tok_s_per_stream": 41.1, "peak_prefill_tok_s": 357665.3, "peak_vram_gb": null, "point_count": 4 }
{ "family": "lfm", "id": "lfm2-5-2-6b", "name": "LFM2.5-2.6B" }
{ "format": "safetensors", "id": "liquidai-lfm2-5-2-6b--bf16", "precision": "BF16", "repository": "LiquidAI/LFM2.5-2.6B" }
lfm25-26b-bf16-rtx4000ada-sglang-tp1
lfm25-26b-bf16-rtx4000ada-sglang-tp1-sweep
2026-08-24T00:00:00
{ "name": "sglang", "version": "0.0.0.dev1+geec794bce.d20260823" }
{ "count": 1, "id": "rtx-4090-24gb", "memory_gb": 24, "name": "GeForce RTX 4090", "total_memory_gb": 24, "vendor": "nvidia" }
lfm25-26b-bf16-rtx4090-sglang-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-4090-24gb", "model": "/api/v1/models/lfm2-5-2-6b", "recipe": "/api/v1/recipes/lfm25-26b-bf16-rtx4090-sglang-tp1", "speed_sweep": "/api/v1/speed-sweeps/lfm25-26b-bf16-rtx4090-sglang-tp1-sweep" }
2026-08-24
{ "best_ttft_ms": 206.6439, "max_concurrency": 4, "max_context_tokens": 130560, "peak_decode_tok_s": null, "peak_decode_tok_s_per_stream": 98.7475, "peak_prefill_tok_s": 583241.8548, "peak_vram_gb": null, "point_count": 6 }
{ "family": "lfm", "id": "lfm2-5-2-6b", "name": "LFM2.5-2.6B" }
{ "format": "safetensors", "id": "liquidai-lfm2-5-2-6b--bf16", "precision": "BF16", "repository": "LiquidAI/LFM2.5-2.6B" }
lfm25-26b-bf16-rtx4090-sglang-tp1
lfm25-26b-bf16-rtx4090-sglang-tp1-sweep
2026-08-23T00:00:00
{ "name": "sglang", "version": "0.0.0.dev1+geec794bce" }
{ "count": 1, "id": "rtx-5090-32gb", "memory_gb": 32, "name": "GeForce RTX 5090", "total_memory_gb": 32, "vendor": "nvidia" }
lfm25-26b-bf16-rtx5090-sglang-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-5090-32gb", "model": "/api/v1/models/lfm2-5-2-6b", "recipe": "/api/v1/recipes/lfm25-26b-bf16-rtx5090-sglang-tp1", "speed_sweep": "/api/v1/speed-sweeps/lfm25-26b-bf16-rtx5090-sglang-tp1-sweep" }
2026-08-23
{ "best_ttft_ms": 24.4, "max_concurrency": 4, "max_context_tokens": 130560, "peak_decode_tok_s": null, "peak_decode_tok_s_per_stream": 197.6, "peak_prefill_tok_s": 819807.3, "peak_vram_gb": null, "point_count": 27 }
{ "family": "lfm", "id": "lfm2-5-2-6b", "name": "LFM2.5-2.6B" }
{ "format": "safetensors", "id": "liquidai-lfm2-5-2-6b--bf16", "precision": "BF16", "repository": "LiquidAI/LFM2.5-2.6B" }
lfm25-26b-bf16-rtx5090-sglang-tp1
lfm25-26b-bf16-rtx5090-sglang-tp1-sweep
2026-08-24T00:00:00
{ "name": "sglang", "version": "0.0.0.dev1+geec794bce" }
{ "count": 1, "id": "rtx-pro-4000-blackwell-24gb", "memory_gb": 24, "name": "NVIDIA RTX PRO 4000 Blackwell", "total_memory_gb": 24, "vendor": "nvidia" }
lfm25-26b-bf16-rtxpro4000-sglang-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-4000-blackwell-24gb", "model": "/api/v1/models/lfm2-5-2-6b", "recipe": "/api/v1/recipes/lfm25-26b-bf16-rtxpro4000-sglang-tp1", "speed_sweep": "/api/v1/speed-sweeps/lfm25-26b-bf16-rtxpro4000-sglang-tp1-sweep" }
2026-08-24
{ "best_ttft_ms": 283.5968, "max_concurrency": 4, "max_context_tokens": 130560, "peak_decode_tok_s": null, "peak_decode_tok_s_per_stream": 71.2735, "peak_prefill_tok_s": 436348.0309, "peak_vram_gb": null, "point_count": 6 }
{ "family": "lfm", "id": "lfm2-5-2-6b", "name": "LFM2.5-2.6B" }
{ "format": "safetensors", "id": "liquidai-lfm2-5-2-6b--bf16", "precision": "BF16", "repository": "LiquidAI/LFM2.5-2.6B" }
lfm25-26b-bf16-rtxpro4000-sglang-tp1
lfm25-26b-bf16-rtxpro4000-sglang-tp1-sweep
2026-08-24T00:00:00
{ "name": "sglang", "version": "0.0.0.dev1+geec794bce" }
{ "count": 1, "id": "rtx-pro-4500-blackwell-32gb", "memory_gb": 32, "name": "NVIDIA RTX PRO 4500 Blackwell", "total_memory_gb": 32, "vendor": "nvidia" }
lfm25-26b-bf16-rtxpro4500-sglang-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-4500-blackwell-32gb", "model": "/api/v1/models/lfm2-5-2-6b", "recipe": "/api/v1/recipes/lfm25-26b-bf16-rtxpro4500-sglang-tp1", "speed_sweep": "/api/v1/speed-sweeps/lfm25-26b-bf16-rtxpro4500-sglang-tp1-sweep" }
2026-08-24
{ "best_ttft_ms": 220.4077, "max_concurrency": 4, "max_context_tokens": 130560, "peak_decode_tok_s": null, "peak_decode_tok_s_per_stream": 92.4287, "peak_prefill_tok_s": 592661.887, "peak_vram_gb": null, "point_count": 6 }
{ "family": "lfm", "id": "lfm2-5-2-6b", "name": "LFM2.5-2.6B" }
{ "format": "safetensors", "id": "liquidai-lfm2-5-2-6b--bf16", "precision": "BF16", "repository": "LiquidAI/LFM2.5-2.6B" }
lfm25-26b-bf16-rtxpro4500-sglang-tp1
lfm25-26b-bf16-rtxpro4500-sglang-tp1-sweep
2026-08-24T00:00:00
{ "name": "sglang", "version": "0.0.0.dev1+geec794bce" }
{ "count": 1, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 96, "vendor": "nvidia" }
lfm25-26b-bf16-rtxpro6000-sglang-tp1-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/lfm2-5-2-6b", "recipe": "/api/v1/recipes/lfm25-26b-bf16-rtxpro6000-sglang-tp1", "speed_sweep": "/api/v1/speed-sweeps/lfm25-26b-bf16-rtxpro6000-sglang-tp1-sweep" }
2026-08-24
{ "best_ttft_ms": 412, "max_concurrency": 4, "max_context_tokens": 130560, "peak_decode_tok_s": null, "peak_decode_tok_s_per_stream": 174.6, "peak_prefill_tok_s": 315897.6, "peak_vram_gb": null, "point_count": 6 }
{ "family": "lfm", "id": "lfm2-5-2-6b", "name": "LFM2.5-2.6B" }
{ "format": "safetensors", "id": "liquidai-lfm2-5-2-6b--bf16", "precision": "BF16", "repository": "LiquidAI/LFM2.5-2.6B" }
lfm25-26b-bf16-rtxpro6000-sglang-tp1
lfm25-26b-bf16-rtxpro6000-sglang-tp1-sweep
null
{ "name": "vllm", "version": "0.20.1-local" }
{ "count": 4, "id": "intel-arc-pro-b70-32gb", "memory_gb": 32, "name": "Intel Arc Pro B70", "total_memory_gb": 128, "vendor": "intel" }
minimax-m2-7-autoround-int4-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-sweep
{ "hardware": "/api/v1/hardware/intel-arc-pro-b70-32gb", "model": "/api/v1/models/minimax-m2-7", "recipe": "/api/v1/recipes/minimax-m2-7-autoround-int4-w4a16-intel-arc-pro-b70-32gb-vllm-tp4", "speed_sweep": "/api/v1/speed-sweeps/minimax-m2-7-autoround-int4-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-sweep" }
2026-05-21
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 93.4436231838881, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "minimax", "id": "minimax-m2-7", "name": "MiniMax-M2.7" }
{ "format": "safetensors", "id": "lasimeri-minimax-m2-7-int4-autoround--autoround-int4-w4a16", "precision": "AutoRound INT4 W4A16", "repository": "Lasimeri/MiniMax-M2.7-int4-AutoRound" }
minimax-m2-7-autoround-int4-w4a16-intel-arc-pro-b70-32gb-vllm-tp4
minimax-m2-7-autoround-int4-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-sweep
null
{ "name": "vllm", "version": "0.20.1-local" }
{ "count": 4, "id": "intel-arc-pro-b70-32gb", "memory_gb": 32, "name": "Intel Arc Pro B70", "total_memory_gb": 128, "vendor": "intel" }
minimax-m2-7-autoround-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-sweep
{ "hardware": "/api/v1/hardware/intel-arc-pro-b70-32gb", "model": "/api/v1/models/minimax-m2-7", "recipe": "/api/v1/recipes/minimax-m2-7-autoround-w4a16-intel-arc-pro-b70-32gb-vllm-tp4", "speed_sweep": "/api/v1/speed-sweeps/minimax-m2-7-autoround-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-sweep" }
2026-05-14
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 61.75284781115523, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "minimax", "id": "minimax-m2-7", "name": "MiniMax-M2.7" }
{ "format": "safetensors", "id": "lasimeri-minimax-m2-7-int4-autoround--autoround-w4a16", "precision": "AutoRound W4A16", "repository": "Lasimeri/MiniMax-M2.7-int4-AutoRound" }
minimax-m2-7-autoround-w4a16-intel-arc-pro-b70-32gb-vllm-tp4
minimax-m2-7-autoround-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-sweep
null
{ "name": "vllm", "version": "0.20.1-local" }
{ "count": 4, "id": "intel-arc-pro-b70-32gb", "memory_gb": 32, "name": "Intel Arc Pro B70", "total_memory_gb": 128, "vendor": "intel" }
minimax-m2-7-int4-autoround-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-sweep
{ "hardware": "/api/v1/hardware/intel-arc-pro-b70-32gb", "model": "/api/v1/models/minimax-m2-7", "recipe": "/api/v1/recipes/minimax-m2-7-int4-autoround-w4a16-intel-arc-pro-b70-32gb-vllm-tp4", "speed_sweep": "/api/v1/speed-sweeps/minimax-m2-7-int4-autoround-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-sweep" }
2026-05-22
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 4096, "peak_decode_tok_s": 94.46, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "minimax", "id": "minimax-m2-7", "name": "MiniMax-M2.7" }
{ "format": "safetensors", "id": "lasimeri-minimax-m2-7-int4-autoround--int4-autoround-w4a16", "precision": "INT4 AutoRound W4A16", "repository": "Lasimeri/MiniMax-M2.7-int4-AutoRound" }
minimax-m2-7-int4-autoround-w4a16-intel-arc-pro-b70-32gb-vllm-tp4
minimax-m2-7-int4-autoround-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-sweep
null
{ "name": "mlx", "version": null }
{ "count": 1, "id": "apple-max-128gb", "memory_gb": 128, "name": "Apple Max 128GB, generation unspecified", "total_memory_gb": 128, "vendor": "apple" }
minimax-m2-7-jangtq-apple-max-128gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-max-128gb", "model": "/api/v1/models/minimax-m2-7", "recipe": "/api/v1/recipes/minimax-m2-7-jangtq-apple-max-128gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/minimax-m2-7-jangtq-apple-max-128gb-mlx-tp1-sweep" }
2026-05-04
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 2048, "peak_decode_tok_s": 31, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "minimax", "id": "minimax-m2-7", "name": "MiniMax-M2.7" }
{ "format": "MLX", "id": "dealignai-minimax-m2-7-jangtq-crack--jangtq", "precision": "JANGTQ", "repository": "dealignai/MiniMax-M2.7-JANGTQ-CRACK" }
minimax-m2-7-jangtq-apple-max-128gb-mlx-tp1
minimax-m2-7-jangtq-apple-max-128gb-mlx-tp1-sweep
null
{ "name": "mlx", "version": null }
{ "count": 1, "id": "apple-m3-ultra-512gb", "memory_gb": 512, "name": "Apple M3 Ultra 512GB", "total_memory_gb": 512, "vendor": "apple" }
minimax-m3-3bit-apple-m3-ultra-512gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-m3-ultra-512gb", "model": "/api/v1/models/minimax-m3", "recipe": "/api/v1/recipes/minimax-m3-3bit-apple-m3-ultra-512gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/minimax-m3-3bit-apple-m3-ultra-512gb-mlx-tp1-sweep" }
2026-06-30
{ "best_ttft_ms": 500, "max_concurrency": 1, "max_context_tokens": 40960000, "peak_decode_tok_s": 21.2, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "minimax", "id": "minimax-m3", "name": "MiniMax-M3" }
{ "format": "MLX", "id": "pipenetwork-minimax-m3-mlx-3bit--3bit", "precision": "3bit", "repository": "pipenetwork/MiniMax-M3-MLX-3bit" }
minimax-m3-3bit-apple-m3-ultra-512gb-mlx-tp1
minimax-m3-3bit-apple-m3-ultra-512gb-mlx-tp1-sweep
2026-08-24T00:00:00
{ "name": "vllm", "version": "source-built patched image; base commit unreported" }
{ "count": 4, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 384, "vendor": "nvidia" }
minimax-m3-mxfp4-rtxpro6000-vllm-tp4-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/minimax-m3", "recipe": "/api/v1/recipes/minimax-m3-mxfp4-rtxpro6000-vllm-tp4", "speed_sweep": "/api/v1/speed-sweeps/minimax-m3-mxfp4-rtxpro6000-vllm-tp4-sweep" }
2026-08-24
{ "best_ttft_ms": null, "max_concurrency": 4, "max_context_tokens": 131072, "peak_decode_tok_s": 184, "peak_decode_tok_s_per_stream": 113, "peak_prefill_tok_s": 2800, "peak_vram_gb": null, "point_count": 3 }
{ "family": "minimax", "id": "minimax-m3", "name": "MiniMax-M3" }
{ "format": "compressed-tensors safetensors", "id": "olka-fi-minimax-m3-mxfp4--mxfp4", "precision": "MXFP4", "repository": "olka-fi/MiniMax-M3-MXFP4" }
minimax-m3-mxfp4-rtxpro6000-vllm-tp4
minimax-m3-mxfp4-rtxpro6000-vllm-tp4-sweep
2026-08-24T00:00:00
{ "name": "vllm", "version": "0.22.0" }
{ "count": 4, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 384, "vendor": "nvidia" }
nemotron-3-ultra-modelopt-mixed-rtxpro6000-vllm-tp4-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/nemotron-3-ultra", "recipe": "/api/v1/recipes/nemotron-3-ultra-modelopt-mixed-rtxpro6000-vllm-tp4", "speed_sweep": "/api/v1/speed-sweeps/nemotron-3-ultra-modelopt-mixed-rtxpro6000-vllm-tp4-sweep" }
2026-06-11
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 261056, "peak_decode_tok_s": 120.7, "peak_decode_tok_s_per_stream": 120.7, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 6 }
{ "family": "nemotron", "id": "nemotron-3-ultra", "name": "Nemotron-3-Ultra" }
{ "format": "ModelOpt mixed quantization", "id": "nvidia-nvidia-nemotron-3-ultra-550b-a55b-nvfp4--nvfp4-mixed", "precision": "NVFP4 mixed", "repository": "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4" }
nemotron-3-ultra-modelopt-mixed-rtxpro6000-vllm-tp4
nemotron-3-ultra-modelopt-mixed-rtxpro6000-vllm-tp4-sweep
null
{ "name": "mlx", "version": "0.31.2" }
{ "count": 1, "id": "apple-m3-ultra-512gb", "memory_gb": 512, "name": "Apple M3 Ultra 512GB", "total_memory_gb": 512, "vendor": "apple" }
nemotron-cascade-2-30b-a3b-4bit-apple-m3-ultra-512gb-mlx-tp1-sweep
{ "hardware": "/api/v1/hardware/apple-m3-ultra-512gb", "model": "/api/v1/models/nemotron-cascade-2-30b-a3b", "recipe": "/api/v1/recipes/nemotron-cascade-2-30b-a3b-4bit-apple-m3-ultra-512gb-mlx-tp1", "speed_sweep": "/api/v1/speed-sweeps/nemotron-cascade-2-30b-a3b-4bit-apple-m3-ultra-512gb-mlx-tp1-sweep" }
2026-04-26
{ "best_ttft_ms": 1308, "max_concurrency": 1, "max_context_tokens": 8192, "peak_decode_tok_s": 140.8, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 1 }
{ "family": "nemotron", "id": "nemotron-cascade-2-30b-a3b", "name": "Nemotron-Cascade-2-30B-A3B" }
{ "format": "MLX", "id": "nvidia-nemotron-cascade-2-30b-a3b--4bit", "precision": "4bit", "repository": "nvidia/Nemotron-Cascade-2-30B-A3B" }
nemotron-cascade-2-30b-a3b-4bit-apple-m3-ultra-512gb-mlx-tp1
nemotron-cascade-2-30b-a3b-4bit-apple-m3-ultra-512gb-mlx-tp1-sweep
2026-08-24T00:00:00
{ "name": "sglang", "version": "0.5.12 fork" }
{ "count": 4, "id": "rtx-pro-6000-blackwell-96gb", "memory_gb": 96, "name": "RTX PRO 6000 Blackwell", "total_memory_gb": 384, "vendor": "nvidia" }
nex-n2-pro-nvfp4-rtxpro6000-sglang-tp4-sweep
{ "hardware": "/api/v1/hardware/rtx-pro-6000-blackwell-96gb", "model": "/api/v1/models/nex-n2-pro", "recipe": "/api/v1/recipes/nex-n2-pro-nvfp4-rtxpro6000-sglang-tp4", "speed_sweep": "/api/v1/speed-sweeps/nex-n2-pro-nvfp4-rtxpro6000-sglang-tp4-sweep" }
2026-06-11
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": null, "peak_decode_tok_s": 93.6, "peak_decode_tok_s_per_stream": 93.6, "peak_prefill_tok_s": null, "peak_vram_gb": null, "point_count": 4 }
{ "family": "nex", "id": "nex-n2-pro", "name": "Nex-N2-Pro" }
{ "format": "ModelOpt", "id": "local-nex-n2-pro-nvfp4--nvfp4", "precision": "NVFP4", "repository": "local/Nex-N2-Pro-NVFP4" }
nex-n2-pro-nvfp4-rtxpro6000-sglang-tp4
nex-n2-pro-nvfp4-rtxpro6000-sglang-tp4-sweep
null
{ "name": "vllm", "version": "v0.26.1rc1.dev668+g3ee2df303 (XPU)" }
{ "count": 1, "id": "intel-arc-pro-b70-32gb", "memory_gb": 32, "name": "Intel Arc Pro B70", "total_memory_gb": 32, "vendor": "intel" }
nvidia-nemotron-3-5-gptq-int4-g64-sym-local-dflash-bf16-local-intel-arc-pro-b70-32gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/intel-arc-pro-b70-32gb", "model": "/api/v1/models/nvidia-nemotron-3-5", "recipe": "/api/v1/recipes/nvidia-nemotron-3-5-gptq-int4-g64-sym-local-dflash-bf16-local-intel-arc-pro-b70-32gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/nvidia-nemotron-3-5-gptq-int4-g64-sym-local-dfla...
2026-08-13
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 16384, "peak_decode_tok_s": 186.6, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 7160, "peak_vram_gb": null, "point_count": 1 }
{ "family": "nemotron", "id": "nvidia-nemotron-3-5", "name": "NVIDIA-Nemotron-3.5" }
{ "format": "safetensors", "id": "nvidia-nvidia-nemotron-3-5-lightning-30b-a3b-bf16--gptq-int4-g64-sym-local-dflash-bf16-local", "precision": "GPTQ-INT4-G64-sym-local+DFlash-BF16-local", "repository": "nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16" }
nvidia-nemotron-3-5-gptq-int4-g64-sym-local-dflash-bf16-local-intel-arc-pro-b70-32gb-vllm-tp1
nvidia-nemotron-3-5-gptq-int4-g64-sym-local-dflash-bf16-local-intel-arc-pro-b70-32gb-vllm-tp1-sweep
null
{ "name": "vllm", "version": "v0.26.1rc1.dev668+g3ee2df303 (XPU)" }
{ "count": 1, "id": "intel-arc-pro-b70-32gb", "memory_gb": 32, "name": "Intel Arc Pro B70", "total_memory_gb": 32, "vendor": "intel" }
nvidia-nemotron-3-5-gptq-int4-g64-sym-local-intel-arc-pro-b70-32gb-vllm-tp1-sweep
{ "hardware": "/api/v1/hardware/intel-arc-pro-b70-32gb", "model": "/api/v1/models/nvidia-nemotron-3-5", "recipe": "/api/v1/recipes/nvidia-nemotron-3-5-gptq-int4-g64-sym-local-intel-arc-pro-b70-32gb-vllm-tp1", "speed_sweep": "/api/v1/speed-sweeps/nvidia-nemotron-3-5-gptq-int4-g64-sym-local-intel-arc-pro-b70-32gb...
2026-08-12
{ "best_ttft_ms": null, "max_concurrency": 1, "max_context_tokens": 16384, "peak_decode_tok_s": 93, "peak_decode_tok_s_per_stream": null, "peak_prefill_tok_s": 8368, "peak_vram_gb": null, "point_count": 1 }
{ "family": "nemotron", "id": "nvidia-nemotron-3-5", "name": "NVIDIA-Nemotron-3.5" }
{ "format": "safetensors", "id": "nvidia-nvidia-nemotron-3-5-lightning-30b-a3b-bf16--gptq-int4-g64-sym-local", "precision": "GPTQ-INT4-G64-sym-local", "repository": "nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16" }
nvidia-nemotron-3-5-gptq-int4-g64-sym-local-intel-arc-pro-b70-32gb-vllm-tp1
nvidia-nemotron-3-5-gptq-int4-g64-sym-local-intel-arc-pro-b70-32gb-vllm-tp1-sweep
End of preview. Expand in Data Studio

Local AI Registry

A source-backed graph of local AI hardware, models, model artifacts, launch recipes, regional prices, raw speed sweeps, and comparable benchmark summaries.

The GitHub monorepo is the source of truth. This dataset repository is a generated, versioned mirror for discovery, analysis, and use with the Hugging Face Datasets library. The live registry and read-only API resolve the same graph.

Tables

Config Primary key Connects through
hardware id Recipe hardware_id; price hardware[].id
models id Model instance model_id
model-instances id Recipe model_instance_id
recipes id Hardware, model instance, and speed_sweeps_ids
prices id Hardware IDs and regional observations
speed-sweeps id Recipe recipe_id and raw measured rows
benchmarks id A comparable projection of one raw speed sweep

Benchmarks do not replace evidence. Each benchmark retains sweep_id, recipe_id, and links to the canonical records from which it was derived.

Use

from datasets import load_dataset

benchmarks = load_dataset(
    "0xSero/local-ai-registry",
    "benchmarks",
    split="train",
)

Every configuration is independently loadable. The JSONL files preserve nested registry fields; null values are explicit unknowns rather than invented defaults.

Snapshot and schemas

  • source.json identifies the Git commit and export timestamp.
  • schema/ contains the JSON Schemas and shared TypeScript interfaces.
  • registry-snapshot.tar.gz contains the exact recursive registry tree used to generate the tables.

Contributing

Follow the registry contribution contract. Contributions are validated and merged in GitHub, then exported here. Do not edit the generated Hugging Face tables as an alternate source of truth.

Provenance

Records carry their own source and provenance fields. Coverage varies by record. Candidate prices and recipes remain visibly marked, and benchmark summaries are only as authoritative as their attached raw sweep evidence.

Downloads last month
63

Collection including 0xSero/local-ai-registry