diff --git a/docs/results.json b/docs/results.json index 32da874..2f3bbe4 100644 --- a/docs/results.json +++ b/docs/results.json @@ -1,6 +1,6 @@ { "meta": { - "generated_at": "2026-05-03T15:19:28Z", + "generated_at": "2026-05-12T11:31:47Z", "system_info": { "distro": "Fedora Linux 43 (Workstation Edition)", "kernel": "6.19.9-200.fc43.x86_64", @@ -16,6 +16,10 @@ "hash": "7957de9dc", "number": "8645" }, + { + "hash": "8e1f9d083", + "number": "9112" + }, { "hash": "ab6120cde", "number": "8997" @@ -29,6 +33,7 @@ "rocm-7_2", "rocm-7_2_2", "rocm-7_2_2-pr21344", + "rocm-7_2_3", "rocm6_4_4", "rocm7-nightlies", "vulkan_amdvlk", @@ -1226,6 +1231,174 @@ "number": "8997" } }, + { + "model": "GLM-4.7-Flash-BF16-00001-of-00002", + "model_clean": "GLM-4.7-Flash-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 403.2, + "tps_std": 2.04, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 29.94, + "file_size_gib": 55.79, + "name_params_b": 29.94, + "quant": "BF16", + "log": "results/GLM-4.7-Flash-BF16-00001-of-00002__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "GLM-4.7-Flash-BF16-00001-of-00002", + "model_clean": "GLM-4.7-Flash-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 20.43, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 29.94, + "file_size_gib": 55.79, + "name_params_b": 29.94, + "quant": "BF16", + "log": "results/GLM-4.7-Flash-BF16-00001-of-00002__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "GLM-4.7-Flash-BF16-00001-of-00002", + "model_clean": "GLM-4.7-Flash-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "pp2048 @ d32768", + "tps_mean": 87.63, + "tps_std": 0.21, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 29.94, + "file_size_gib": 55.79, + "name_params_b": 29.94, + "quant": "BF16", + "log": "results/GLM-4.7-Flash-BF16-00001-of-00002__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "GLM-4.7-Flash-BF16-00001-of-00002", + "model_clean": "GLM-4.7-Flash-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "tg32 @ d32768", + "tps_mean": 15.31, + "tps_std": 0.02, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 29.94, + "file_size_gib": 55.79, + "name_params_b": 29.94, + "quant": "BF16", + "log": "results/GLM-4.7-Flash-BF16-00001-of-00002__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "GLM-4.7-Flash-BF16-00001-of-00002", + "model_clean": "GLM-4.7-Flash-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "pp2048 @ d65536", + "tps_mean": 46.45, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 29.94, + "file_size_gib": 55.79, + "name_params_b": 29.94, + "quant": "BF16", + "log": "results/GLM-4.7-Flash-BF16-00001-of-00002__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "GLM-4.7-Flash-BF16-00001-of-00002", + "model_clean": "GLM-4.7-Flash-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "tg32 @ d65536", + "tps_mean": 11.98, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 29.94, + "file_size_gib": 55.79, + "name_params_b": 29.94, + "quant": "BF16", + "log": "results/GLM-4.7-Flash-BF16-00001-of-00002__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, { "model": "GLM-4.7-Flash-BF16-00001-of-00002", "model_clean": "GLM-4.7-Flash-BF16", @@ -2228,6 +2401,174 @@ "number": "8997" } }, + { + "model": "GLM-4.7-Flash-UD-Q8_K_XL", + "model_clean": "GLM-4.7-Flash-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 905.99, + "tps_std": 2.05, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 29.94, + "file_size_gib": 32.7, + "name_params_b": 29.94, + "quant": "Q8_K_XL", + "log": "results/GLM-4.7-Flash-UD-Q8_K_XL__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "GLM-4.7-Flash-UD-Q8_K_XL", + "model_clean": "GLM-4.7-Flash-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 32.65, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 29.94, + "file_size_gib": 32.7, + "name_params_b": 29.94, + "quant": "Q8_K_XL", + "log": "results/GLM-4.7-Flash-UD-Q8_K_XL__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "GLM-4.7-Flash-UD-Q8_K_XL", + "model_clean": "GLM-4.7-Flash-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "pp2048 @ d32768", + "tps_mean": 91.59, + "tps_std": 0.16, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 29.94, + "file_size_gib": 32.7, + "name_params_b": 29.94, + "quant": "Q8_K_XL", + "log": "results/GLM-4.7-Flash-UD-Q8_K_XL__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "GLM-4.7-Flash-UD-Q8_K_XL", + "model_clean": "GLM-4.7-Flash-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "tg32 @ d32768", + "tps_mean": 20.36, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 29.94, + "file_size_gib": 32.7, + "name_params_b": 29.94, + "quant": "Q8_K_XL", + "log": "results/GLM-4.7-Flash-UD-Q8_K_XL__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "GLM-4.7-Flash-UD-Q8_K_XL", + "model_clean": "GLM-4.7-Flash-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "pp2048 @ d65536", + "tps_mean": 47.51, + "tps_std": 0.05, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 29.94, + "file_size_gib": 32.7, + "name_params_b": 29.94, + "quant": "Q8_K_XL", + "log": "results/GLM-4.7-Flash-UD-Q8_K_XL__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "GLM-4.7-Flash-UD-Q8_K_XL", + "model_clean": "GLM-4.7-Flash-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "tg32 @ d65536", + "tps_mean": 14.86, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 29.94, + "file_size_gib": 32.7, + "name_params_b": 29.94, + "quant": "Q8_K_XL", + "log": "results/GLM-4.7-Flash-UD-Q8_K_XL__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, { "model": "GLM-4.7-Flash-UD-Q8_K_XL", "model_clean": "GLM-4.7-Flash-UD-Q8_K_XL", @@ -3112,6 +3453,143 @@ "rpc": false, "build": null }, + { + "model": "MiniMax-M2.7-UD-Q3_K_S-00001-of-00003", + "model_clean": "MiniMax-M2.7-UD-Q3_K_S", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 236.96, + "tps_std": 1.25, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 228.69, + "file_size_gib": 87.2, + "name_params_b": 228.69, + "quant": "Q3_K_S", + "log": "results/MiniMax-M2.7-UD-Q3_K_S-00001-of-00003__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "MiniMax-M2.7-UD-Q3_K_S-00001-of-00003", + "model_clean": "MiniMax-M2.7-UD-Q3_K_S", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 22.81, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 228.69, + "file_size_gib": 87.2, + "name_params_b": 228.69, + "quant": "Q3_K_S", + "log": "results/MiniMax-M2.7-UD-Q3_K_S-00001-of-00003__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "MiniMax-M2.7-UD-Q3_K_S-00001-of-00003", + "model_clean": "MiniMax-M2.7-UD-Q3_K_S", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "pp2048 @ d32768", + "tps_mean": 62.13, + "tps_std": 0.56, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 228.69, + "file_size_gib": 87.2, + "name_params_b": 228.69, + "quant": "Q3_K_S", + "log": "results/MiniMax-M2.7-UD-Q3_K_S-00001-of-00003__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "MiniMax-M2.7-UD-Q3_K_S-00001-of-00003", + "model_clean": "MiniMax-M2.7-UD-Q3_K_S", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "tg32 @ d32768", + "tps_mean": 6.15, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 228.69, + "file_size_gib": 87.2, + "name_params_b": 228.69, + "quant": "Q3_K_S", + "log": "results/MiniMax-M2.7-UD-Q3_K_S-00001-of-00003__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "MiniMax-M2.7-UD-Q3_K_S-00001-of-00003", + "model_clean": "MiniMax-M2.7-UD-Q3_K_S", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": null, + "tps_mean": null, + "tps_std": null, + "error": false, + "error_type": null, + "backend": null, + "ngl": null, + "mmap": null, + "params_b": null, + "file_size_gib": null, + "name_params_b": null, + "quant": "Q3_K_S", + "log": "results/MiniMax-M2.7-UD-Q3_K_S-00001-of-00003__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": null + }, { "model": "MiniMax-M2.7-UD-Q3_K_S-00001-of-00003", "model_clean": "MiniMax-M2.7-UD-Q3_K_S", @@ -11345,6 +11823,174 @@ "number": "8997" } }, + { + "model": "gemma-4-26B-A4B-it-BF16-00001-of-00002", + "model_clean": "gemma-4-26B-A4B-it-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 796.06, + "tps_std": 126.76, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 47.02, + "name_params_b": 25.23, + "quant": "BF16", + "log": "results/gemma-4-26B-A4B-it-BF16-00001-of-00002__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-26B-A4B-it-BF16-00001-of-00002", + "model_clean": "gemma-4-26B-A4B-it-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 22.68, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 47.02, + "name_params_b": 25.23, + "quant": "BF16", + "log": "results/gemma-4-26B-A4B-it-BF16-00001-of-00002__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-26B-A4B-it-BF16-00001-of-00002", + "model_clean": "gemma-4-26B-A4B-it-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "pp2048 @ d32768", + "tps_mean": 628.2, + "tps_std": 5.71, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 47.02, + "name_params_b": 25.23, + "quant": "BF16", + "log": "results/gemma-4-26B-A4B-it-BF16-00001-of-00002__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-26B-A4B-it-BF16-00001-of-00002", + "model_clean": "gemma-4-26B-A4B-it-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "tg32 @ d32768", + "tps_mean": 20.12, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 47.02, + "name_params_b": 25.23, + "quant": "BF16", + "log": "results/gemma-4-26B-A4B-it-BF16-00001-of-00002__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-26B-A4B-it-BF16-00001-of-00002", + "model_clean": "gemma-4-26B-A4B-it-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "pp2048 @ d65536", + "tps_mean": 426.61, + "tps_std": 3.98, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 47.02, + "name_params_b": 25.23, + "quant": "BF16", + "log": "results/gemma-4-26B-A4B-it-BF16-00001-of-00002__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-26B-A4B-it-BF16-00001-of-00002", + "model_clean": "gemma-4-26B-A4B-it-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "tg32 @ d65536", + "tps_mean": 19.55, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 47.02, + "name_params_b": 25.23, + "quant": "BF16", + "log": "results/gemma-4-26B-A4B-it-BF16-00001-of-00002__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, { "model": "gemma-4-26B-A4B-it-BF16-00001-of-00002", "model_clean": "gemma-4-26B-A4B-it-BF16", @@ -12353,6 +12999,174 @@ "number": "8997" } }, + { + "model": "gemma-4-26B-A4B-it-UD-Q4_K_XL", + "model_clean": "gemma-4-26B-A4B-it-UD-Q4_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 1301.66, + "tps_std": 10.69, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 15.9, + "name_params_b": 25.23, + "quant": "Q4_K_XL", + "log": "results/gemma-4-26B-A4B-it-UD-Q4_K_XL__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-26B-A4B-it-UD-Q4_K_XL", + "model_clean": "gemma-4-26B-A4B-it-UD-Q4_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 46.42, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 15.9, + "name_params_b": 25.23, + "quant": "Q4_K_XL", + "log": "results/gemma-4-26B-A4B-it-UD-Q4_K_XL__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-26B-A4B-it-UD-Q4_K_XL", + "model_clean": "gemma-4-26B-A4B-it-UD-Q4_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "pp2048 @ d32768", + "tps_mean": 693.56, + "tps_std": 10.92, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 15.9, + "name_params_b": 25.23, + "quant": "Q4_K_XL", + "log": "results/gemma-4-26B-A4B-it-UD-Q4_K_XL__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-26B-A4B-it-UD-Q4_K_XL", + "model_clean": "gemma-4-26B-A4B-it-UD-Q4_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "tg32 @ d32768", + "tps_mean": 36.83, + "tps_std": 0.15, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 15.9, + "name_params_b": 25.23, + "quant": "Q4_K_XL", + "log": "results/gemma-4-26B-A4B-it-UD-Q4_K_XL__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-26B-A4B-it-UD-Q4_K_XL", + "model_clean": "gemma-4-26B-A4B-it-UD-Q4_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "pp2048 @ d65536", + "tps_mean": 450.75, + "tps_std": 0.77, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 15.9, + "name_params_b": 25.23, + "quant": "Q4_K_XL", + "log": "results/gemma-4-26B-A4B-it-UD-Q4_K_XL__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-26B-A4B-it-UD-Q4_K_XL", + "model_clean": "gemma-4-26B-A4B-it-UD-Q4_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "tg32 @ d65536", + "tps_mean": 34.84, + "tps_std": 0.16, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 15.9, + "name_params_b": 25.23, + "quant": "Q4_K_XL", + "log": "results/gemma-4-26B-A4B-it-UD-Q4_K_XL__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, { "model": "gemma-4-26B-A4B-it-UD-Q4_K_XL", "model_clean": "gemma-4-26B-A4B-it-UD-Q4_K_XL", @@ -13361,6 +14175,174 @@ "number": "8997" } }, + { + "model": "gemma-4-26B-A4B-it-UD-Q8_K_XL", + "model_clean": "gemma-4-26B-A4B-it-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 1301.93, + "tps_std": 18.07, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 25.94, + "name_params_b": 25.23, + "quant": "Q8_K_XL", + "log": "results/gemma-4-26B-A4B-it-UD-Q8_K_XL__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-26B-A4B-it-UD-Q8_K_XL", + "model_clean": "gemma-4-26B-A4B-it-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 41.35, + "tps_std": 0.03, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 25.94, + "name_params_b": 25.23, + "quant": "Q8_K_XL", + "log": "results/gemma-4-26B-A4B-it-UD-Q8_K_XL__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-26B-A4B-it-UD-Q8_K_XL", + "model_clean": "gemma-4-26B-A4B-it-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "pp2048 @ d32768", + "tps_mean": 695.0, + "tps_std": 7.26, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 25.94, + "name_params_b": 25.23, + "quant": "Q8_K_XL", + "log": "results/gemma-4-26B-A4B-it-UD-Q8_K_XL__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-26B-A4B-it-UD-Q8_K_XL", + "model_clean": "gemma-4-26B-A4B-it-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "tg32 @ d32768", + "tps_mean": 33.41, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 25.94, + "name_params_b": 25.23, + "quant": "Q8_K_XL", + "log": "results/gemma-4-26B-A4B-it-UD-Q8_K_XL__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-26B-A4B-it-UD-Q8_K_XL", + "model_clean": "gemma-4-26B-A4B-it-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "pp2048 @ d65536", + "tps_mean": 447.63, + "tps_std": 3.37, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 25.94, + "name_params_b": 25.23, + "quant": "Q8_K_XL", + "log": "results/gemma-4-26B-A4B-it-UD-Q8_K_XL__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-26B-A4B-it-UD-Q8_K_XL", + "model_clean": "gemma-4-26B-A4B-it-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "tg32 @ d65536", + "tps_mean": 31.76, + "tps_std": 0.04, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 25.23, + "file_size_gib": 25.94, + "name_params_b": 25.23, + "quant": "Q8_K_XL", + "log": "results/gemma-4-26B-A4B-it-UD-Q8_K_XL__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, { "model": "gemma-4-26B-A4B-it-UD-Q8_K_XL", "model_clean": "gemma-4-26B-A4B-it-UD-Q8_K_XL", @@ -14369,6 +15351,174 @@ "number": "8997" } }, + { + "model": "gemma-4-31B-it-BF16-00001-of-00002", + "model_clean": "gemma-4-31B-it-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 376.47, + "tps_std": 1.77, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 57.18, + "name_params_b": 30.7, + "quant": "BF16", + "log": "results/gemma-4-31B-it-BF16-00001-of-00002__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-31B-it-BF16-00001-of-00002", + "model_clean": "gemma-4-31B-it-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 3.44, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 57.18, + "name_params_b": 30.7, + "quant": "BF16", + "log": "results/gemma-4-31B-it-BF16-00001-of-00002__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-31B-it-BF16-00001-of-00002", + "model_clean": "gemma-4-31B-it-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "pp2048 @ d32768", + "tps_mean": 152.54, + "tps_std": 2.43, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 57.18, + "name_params_b": 30.7, + "quant": "BF16", + "log": "results/gemma-4-31B-it-BF16-00001-of-00002__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-31B-it-BF16-00001-of-00002", + "model_clean": "gemma-4-31B-it-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "tg32 @ d32768", + "tps_mean": 3.18, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 57.18, + "name_params_b": 30.7, + "quant": "BF16", + "log": "results/gemma-4-31B-it-BF16-00001-of-00002__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-31B-it-BF16-00001-of-00002", + "model_clean": "gemma-4-31B-it-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "pp2048 @ d65536", + "tps_mean": 98.53, + "tps_std": 0.11, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 57.18, + "name_params_b": 30.7, + "quant": "BF16", + "log": "results/gemma-4-31B-it-BF16-00001-of-00002__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-31B-it-BF16-00001-of-00002", + "model_clean": "gemma-4-31B-it-BF16", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "tg32 @ d65536", + "tps_mean": 3.08, + "tps_std": 0.02, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 57.18, + "name_params_b": 30.7, + "quant": "BF16", + "log": "results/gemma-4-31B-it-BF16-00001-of-00002__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, { "model": "gemma-4-31B-it-BF16-00001-of-00002", "model_clean": "gemma-4-31B-it-BF16", @@ -15284,6 +16434,174 @@ "number": "8997" } }, + { + "model": "gemma-4-31B-it-UD-Q4_K_XL", + "model_clean": "gemma-4-31B-it-UD-Q4_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 309.16, + "tps_std": 0.31, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 17.46, + "name_params_b": 30.7, + "quant": "Q4_K_XL", + "log": "results/gemma-4-31B-it-UD-Q4_K_XL__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-31B-it-UD-Q4_K_XL", + "model_clean": "gemma-4-31B-it-UD-Q4_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 10.49, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 17.46, + "name_params_b": 30.7, + "quant": "Q4_K_XL", + "log": "results/gemma-4-31B-it-UD-Q4_K_XL__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-31B-it-UD-Q4_K_XL", + "model_clean": "gemma-4-31B-it-UD-Q4_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "pp2048 @ d32768", + "tps_mean": 139.29, + "tps_std": 1.61, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 17.46, + "name_params_b": 30.7, + "quant": "Q4_K_XL", + "log": "results/gemma-4-31B-it-UD-Q4_K_XL__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-31B-it-UD-Q4_K_XL", + "model_clean": "gemma-4-31B-it-UD-Q4_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "tg32 @ d32768", + "tps_mean": 8.1, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 17.46, + "name_params_b": 30.7, + "quant": "Q4_K_XL", + "log": "results/gemma-4-31B-it-UD-Q4_K_XL__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-31B-it-UD-Q4_K_XL", + "model_clean": "gemma-4-31B-it-UD-Q4_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "pp2048 @ d65536", + "tps_mean": 92.8, + "tps_std": 0.18, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 17.46, + "name_params_b": 30.7, + "quant": "Q4_K_XL", + "log": "results/gemma-4-31B-it-UD-Q4_K_XL__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-31B-it-UD-Q4_K_XL", + "model_clean": "gemma-4-31B-it-UD-Q4_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "tg32 @ d65536", + "tps_mean": 7.51, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 17.46, + "name_params_b": 30.7, + "quant": "Q4_K_XL", + "log": "results/gemma-4-31B-it-UD-Q4_K_XL__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, { "model": "gemma-4-31B-it-UD-Q4_K_XL", "model_clean": "gemma-4-31B-it-UD-Q4_K_XL", @@ -16230,6 +17548,174 @@ "number": "8997" } }, + { + "model": "gemma-4-31B-it-UD-Q8_K_XL", + "model_clean": "gemma-4-31B-it-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 309.61, + "tps_std": 0.81, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 32.6, + "name_params_b": 30.7, + "quant": "Q8_K_XL", + "log": "results/gemma-4-31B-it-UD-Q8_K_XL__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-31B-it-UD-Q8_K_XL", + "model_clean": "gemma-4-31B-it-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 6.15, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 32.6, + "name_params_b": 30.7, + "quant": "Q8_K_XL", + "log": "results/gemma-4-31B-it-UD-Q8_K_XL__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-31B-it-UD-Q8_K_XL", + "model_clean": "gemma-4-31B-it-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "pp2048 @ d32768", + "tps_mean": 139.33, + "tps_std": 1.73, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 32.6, + "name_params_b": 30.7, + "quant": "Q8_K_XL", + "log": "results/gemma-4-31B-it-UD-Q8_K_XL__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-31B-it-UD-Q8_K_XL", + "model_clean": "gemma-4-31B-it-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "tg32 @ d32768", + "tps_mean": 5.25, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 32.6, + "name_params_b": 30.7, + "quant": "Q8_K_XL", + "log": "results/gemma-4-31B-it-UD-Q8_K_XL__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-31B-it-UD-Q8_K_XL", + "model_clean": "gemma-4-31B-it-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "pp2048 @ d65536", + "tps_mean": 93.3, + "tps_std": 0.54, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 32.6, + "name_params_b": 30.7, + "quant": "Q8_K_XL", + "log": "results/gemma-4-31B-it-UD-Q8_K_XL__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gemma-4-31B-it-UD-Q8_K_XL", + "model_clean": "gemma-4-31B-it-UD-Q8_K_XL", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "tg32 @ d65536", + "tps_mean": 5.0, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 30.7, + "file_size_gib": 32.6, + "name_params_b": 30.7, + "quant": "Q8_K_XL", + "log": "results/gemma-4-31B-it-UD-Q8_K_XL__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, { "model": "gemma-4-31B-it-UD-Q8_K_XL", "model_clean": "gemma-4-31B-it-UD-Q8_K_XL", @@ -17176,6 +18662,174 @@ "number": "8997" } }, + { + "model": "gpt-oss-120b-mxfp4-00001-of-00003", + "model_clean": "gpt-oss-120b-mxfp4", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 635.33, + "tps_std": 4.03, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 116.83, + "file_size_gib": 59.02, + "name_params_b": 116.83, + "quant": "MXFP4", + "log": "results/gpt-oss-120b-mxfp4-00001-of-00003__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gpt-oss-120b-mxfp4-00001-of-00003", + "model_clean": "gpt-oss-120b-mxfp4", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 50.99, + "tps_std": 0.05, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 116.83, + "file_size_gib": 59.02, + "name_params_b": 116.83, + "quant": "MXFP4", + "log": "results/gpt-oss-120b-mxfp4-00001-of-00003__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gpt-oss-120b-mxfp4-00001-of-00003", + "model_clean": "gpt-oss-120b-mxfp4", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "pp2048 @ d32768", + "tps_mean": 302.31, + "tps_std": 0.46, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 116.83, + "file_size_gib": 59.02, + "name_params_b": 116.83, + "quant": "MXFP4", + "log": "results/gpt-oss-120b-mxfp4-00001-of-00003__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gpt-oss-120b-mxfp4-00001-of-00003", + "model_clean": "gpt-oss-120b-mxfp4", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "tg32 @ d32768", + "tps_mean": 35.94, + "tps_std": 0.12, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 116.83, + "file_size_gib": 59.02, + "name_params_b": 116.83, + "quant": "MXFP4", + "log": "results/gpt-oss-120b-mxfp4-00001-of-00003__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gpt-oss-120b-mxfp4-00001-of-00003", + "model_clean": "gpt-oss-120b-mxfp4", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "pp2048 @ d65536", + "tps_mean": 175.25, + "tps_std": 0.49, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 116.83, + "file_size_gib": 59.02, + "name_params_b": 116.83, + "quant": "MXFP4", + "log": "results/gpt-oss-120b-mxfp4-00001-of-00003__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gpt-oss-120b-mxfp4-00001-of-00003", + "model_clean": "gpt-oss-120b-mxfp4", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "tg32 @ d65536", + "tps_mean": 27.78, + "tps_std": 0.06, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 116.83, + "file_size_gib": 59.02, + "name_params_b": 116.83, + "quant": "MXFP4", + "log": "results/gpt-oss-120b-mxfp4-00001-of-00003__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, { "model": "gpt-oss-120b-mxfp4-00001-of-00003", "model_clean": "gpt-oss-120b-mxfp4", @@ -18184,6 +19838,174 @@ "number": "8997" } }, + { + "model": "gpt-oss-20b-mxfp4", + "model_clean": "gpt-oss-20b-mxfp4", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 1636.3, + "tps_std": 15.57, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 20.91, + "file_size_gib": 11.27, + "name_params_b": 20.91, + "quant": "MXFP4", + "log": "results/gpt-oss-20b-mxfp4__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gpt-oss-20b-mxfp4", + "model_clean": "gpt-oss-20b-mxfp4", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 72.58, + "tps_std": 0.05, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 20.91, + "file_size_gib": 11.27, + "name_params_b": 20.91, + "quant": "MXFP4", + "log": "results/gpt-oss-20b-mxfp4__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gpt-oss-20b-mxfp4", + "model_clean": "gpt-oss-20b-mxfp4", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "pp2048 @ d32768", + "tps_mean": 487.06, + "tps_std": 3.04, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 20.91, + "file_size_gib": 11.27, + "name_params_b": 20.91, + "quant": "MXFP4", + "log": "results/gpt-oss-20b-mxfp4__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gpt-oss-20b-mxfp4", + "model_clean": "gpt-oss-20b-mxfp4", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "tg32 @ d32768", + "tps_mean": 51.73, + "tps_std": 0.18, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 20.91, + "file_size_gib": 11.27, + "name_params_b": 20.91, + "quant": "MXFP4", + "log": "results/gpt-oss-20b-mxfp4__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gpt-oss-20b-mxfp4", + "model_clean": "gpt-oss-20b-mxfp4", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "pp2048 @ d65536", + "tps_mean": 276.26, + "tps_std": 0.34, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 20.91, + "file_size_gib": 11.27, + "name_params_b": 20.91, + "quant": "MXFP4", + "log": "results/gpt-oss-20b-mxfp4__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "gpt-oss-20b-mxfp4", + "model_clean": "gpt-oss-20b-mxfp4", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "tg32 @ d65536", + "tps_mean": 40.32, + "tps_std": 0.08, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 20.91, + "file_size_gib": 11.27, + "name_params_b": 20.91, + "quant": "MXFP4", + "log": "results/gpt-oss-20b-mxfp4__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, { "model": "gpt-oss-20b-mxfp4", "model_clean": "gpt-oss-20b-mxfp4", @@ -19192,6 +21014,174 @@ "number": "8997" } }, + { + "model": "llama-2-7b.Q4_0", + "model_clean": "llama-2-7b.Q4_0", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 1543.35, + "tps_std": 8.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 6.74, + "file_size_gib": 3.56, + "name_params_b": 6.74, + "quant": "Q4_0", + "log": "results/llama-2-7b.Q4_0__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "llama-2-7b.Q4_0", + "model_clean": "llama-2-7b.Q4_0", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 50.5, + "tps_std": 0.12, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 6.74, + "file_size_gib": 3.56, + "name_params_b": 6.74, + "quant": "Q4_0", + "log": "results/llama-2-7b.Q4_0__rocm-7_2_3__fa1.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "llama-2-7b.Q4_0", + "model_clean": "llama-2-7b.Q4_0", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "pp2048 @ d32768", + "tps_mean": 159.17, + "tps_std": 0.71, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 6.74, + "file_size_gib": 3.56, + "name_params_b": 6.74, + "quant": "Q4_0", + "log": "results/llama-2-7b.Q4_0__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "llama-2-7b.Q4_0", + "model_clean": "llama-2-7b.Q4_0", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx32768", + "context_tokens": 32768, + "test": "tg32 @ d32768", + "tps_mean": 5.64, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 6.74, + "file_size_gib": 3.56, + "name_params_b": 6.74, + "quant": "Q4_0", + "log": "results/llama-2-7b.Q4_0__rocm-7_2_3__fa1__longctx32768.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "llama-2-7b.Q4_0", + "model_clean": "llama-2-7b.Q4_0", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "pp2048 @ d65536", + "tps_mean": 72.03, + "tps_std": 0.23, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 6.74, + "file_size_gib": 3.56, + "name_params_b": 6.74, + "quant": "Q4_0", + "log": "results/llama-2-7b.Q4_0__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, + { + "model": "llama-2-7b.Q4_0", + "model_clean": "llama-2-7b.Q4_0", + "env": "rocm-7_2_3", + "env_base": "rocm", + "env_variant": "7_2_3", + "fa": true, + "context": "longctx65536", + "context_tokens": 65536, + "test": "tg32 @ d65536", + "tps_mean": 3.0, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": 0, + "params_b": 6.74, + "file_size_gib": 3.56, + "name_params_b": 6.74, + "quant": "Q4_0", + "log": "results/llama-2-7b.Q4_0__rocm-7_2_3__fa1__longctx65536.log", + "rpc": false, + "build": { + "hash": "8e1f9d083", + "number": "9112" + } + }, { "model": "llama-2-7b.Q4_0", "model_clean": "llama-2-7b.Q4_0",