Devstral-2-123B-Instruct-2512 on AMD GPUs

115 results · 3 sources · 142 measurements

Grouped by reported model identity. Exact weights, quantizations and software can differ. How results are reviewed

Documented results

24 selected tests


View ReportCustom evaluation	Devstral-2-123B-Instruct-2512Q4_K_MModel ref. ↗Conflicting evidence	HumanEval100 problems	88 %Reported pass@1 · n = 100	—	7.05 tok/sGeneration · mean · per request	2 × W7800Layer split · 48 GB VRAM/GPU Configuration Memory 48 GB VRAM/GPU OS Ubuntu 24.04 LTS Engine Ollama Backend ROCm Harness BoJl unified evaluation harness (2026-08) Prompt / output ? / ? tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View Reportllama-bench	Devstral-2-123B-Instruct-2512Q3_K_SModel ref. ↗	Token speed512 input tokens · 0 existing KV tokens	—	133.82 tok/sPrefill · mean · per request	—	2 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Linux (inferred; distribution unreported) (inferred) Engine llama.cpp Backend vulkan_amdvlk Harness llama-bench Prompt / output 512 / ? tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View Reportllama-bench	Devstral-2-123B-Instruct-2512Q3_K_SModel ref. ↗	Token speed512 input tokens · 0 existing KV tokens	—	43.07 tok/sPrefill · mean · per request	—	2 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Fedora 42 — Linux 6.15.9-201.fc42.x86_64 (Sat Aug 2 11:37:34 UTC 2025) Engine llama.cpp Backend vulkan_amdvlk Harness llama-bench Prompt / output 512 / ? tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View ReportCustom evaluation	Devstral-2-123B-Instruct-2512Q4_K_MModel ref. ↗Conflicting evidence	Token speedsynth-bulk-dispatcher-impl	—	—	7 tok/sGeneration · mean · per request	2 × W7800Layer split · 48 GB VRAM/GPU Configuration Memory 48 GB VRAM/GPU OS Ubuntu 24.04 LTS Engine Ollama Backend ROCm Harness BoJl unified evaluation harness (2026-08) Prompt / output ? / 729 tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View Reportllama-bench	Devstral-2-123B-Instruct-2512Q3_K_SModel ref. ↗	Token speed128 output tokens · 0 existing KV tokens	—	—	7.55 tok/sGeneration · mean · per request	2 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Linux (inferred; distribution unreported) (inferred) Engine llama.cpp Backend vulkan_amdvlk Harness llama-bench Prompt / output ? / 128 tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View Reportllama-bench	Devstral-2-123B-Instruct-2512Q3_K_SModel ref. ↗	Token speed128 output tokens · 0 existing KV tokens	—	—	7.91 tok/sGeneration · mean · per request	2 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Fedora 42 — Linux 6.15.9-201.fc42.x86_64 (Sat Aug 2 11:37:34 UTC 2025) Engine llama.cpp Backend vulkan_amdvlk Harness llama-bench Prompt / output ? / 128 tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View ReportCustom evaluation	Devstral-2-123B-Instruct-2512Q4_K_MModel ref. ↗Conflicting evidence	Token speedsynth-horner-mpmath-derivative	—	—	7.06 tok/sGeneration · mean · per request	2 × W7800Layer split · 48 GB VRAM/GPU Configuration Memory 48 GB VRAM/GPU OS Ubuntu 24.04 LTS Engine Ollama Backend ROCm Harness BoJl unified evaluation harness (2026-08) Prompt / output ? / 168 tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View Reportllama-bench	Devstral-2-123B-Instruct-2512Q3_K_SModel ref. ↗	Token speed2,048 input tokens · 16,384 existing KV tokens	—	44.89 tok/sPrefill · mean · per request	—	2 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Linux (inferred; distribution unreported) (inferred) Engine llama.cpp Backend vulkan_amdvlk Harness llama-bench Prompt / output 2048 / ? tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View Reportllama-bench	Devstral-2-123B-Instruct-2512Q3_K_SModel ref. ↗	Token speed2,048 input tokens · 16,384 existing KV tokens	—	59.09 tok/sPrefill · mean · per request	—	2 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Fedora 42 — Linux 6.15.9-201.fc42.x86_64 (Sat Aug 2 11:37:34 UTC 2025) Engine llama.cpp Backend vulkan_amdvlk Harness llama-bench Prompt / output 2048 / ? tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View ReportCustom evaluation	Devstral-2-123B-Instruct-2512Q4_K_MModel ref. ↗Conflicting evidence	Token speedsynth-invoke-sign-artifacts-ps	—	—	6.99 tok/sGeneration · mean · per request	2 × W7800Layer split · 48 GB VRAM/GPU Configuration Memory 48 GB VRAM/GPU OS Ubuntu 24.04 LTS Engine Ollama Backend ROCm Harness BoJl unified evaluation harness (2026-08) Prompt / output ? / 753 tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View Reportllama-bench	Devstral-2-123B-Instruct-2512Q3_K_SModel ref. ↗	Token speed32 output tokens · 16,384 existing KV tokens	—	—	4.37 tok/sGeneration · mean · per request	2 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Linux (inferred; distribution unreported) (inferred) Engine llama.cpp Backend vulkan_amdvlk Harness llama-bench Prompt / output ? / 32 tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View Reportllama-bench	Devstral-2-123B-Instruct-2512Q3_K_SModel ref. ↗	Token speed32 output tokens · 16,384 existing KV tokens	—	—	7.45 tok/sGeneration · mean · per request	2 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Fedora 42 — Linux 6.15.9-201.fc42.x86_64 (Sat Aug 2 11:37:34 UTC 2025) Engine llama.cpp Backend vulkan_amdvlk Harness llama-bench Prompt / output ? / 32 tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View ReportCustom evaluation	Devstral-2-123B-Instruct-2512Q4_K_MModel ref. ↗Conflicting evidence	Token speedsynth-legacy-rename-toolchain	—	—	7.04 tok/sGeneration · mean · per request	2 × W7800Layer split · 48 GB VRAM/GPU Configuration Memory 48 GB VRAM/GPU OS Ubuntu 24.04 LTS Engine Ollama Backend ROCm Harness BoJl unified evaluation harness (2026-08) Prompt / output ? / 859 tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View Reportllama-bench	Devstral-2-123B-Instruct-2512Q3_K_SModel ref. ↗	Token speed2,048 input tokens · 32,768 existing KV tokens	—	16.45 tok/sPrefill · mean · per request	—	2 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Linux (inferred; distribution unreported) (inferred) Engine llama.cpp Backend vulkan_amdvlk Harness llama-bench Prompt / output 2048 / ? tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View Reportllama-bench	Devstral-2-123B-Instruct-2512Q3_K_SModel ref. ↗	Token speed2,048 input tokens · 32,768 existing KV tokens	—	53.36 tok/sPrefill · mean · per request	—	2 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Fedora 42 — Linux 6.15.9-201.fc42.x86_64 (Sat Aug 2 11:37:34 UTC 2025) Engine llama.cpp Backend vulkan_amdvlk Harness llama-bench Prompt / output 2048 / ? tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View ReportCustom evaluation	Devstral-2-123B-Instruct-2512Q4_K_MModel ref. ↗Conflicting evidence	Token speedsynth-nuget-multitarget-design	—	—	7.05 tok/sGeneration · mean · per request	2 × W7800Layer split · 48 GB VRAM/GPU Configuration Memory 48 GB VRAM/GPU OS Ubuntu 24.04 LTS Engine Ollama Backend ROCm Harness BoJl unified evaluation harness (2026-08) Prompt / output ? / 329 tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View Reportllama-bench	Devstral-2-123B-Instruct-2512Q3_K_SModel ref. ↗	Token speed32 output tokens · 32,768 existing KV tokens	—	—	3.13 tok/sGeneration · mean · per request	2 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Linux (inferred; distribution unreported) (inferred) Engine llama.cpp Backend vulkan_amdvlk Harness llama-bench Prompt / output ? / 32 tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View Reportllama-bench	Devstral-2-123B-Instruct-2512Q3_K_SModel ref. ↗	Token speed32 output tokens · 32,768 existing KV tokens	—	—	6.99 tok/sGeneration · mean · per request	2 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Fedora 42 — Linux 6.15.9-201.fc42.x86_64 (Sat Aug 2 11:37:34 UTC 2025) Engine llama.cpp Backend vulkan_amdvlk Harness llama-bench Prompt / output ? / 32 tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View ReportCustom evaluation	Devstral-2-123B-Instruct-2512Q4_K_MModel ref. ↗Conflicting evidence	Token speedsynth-parser-nunit-tests	—	—	7.04 tok/sGeneration · mean · per request	2 × W7800Layer split · 48 GB VRAM/GPU Configuration Memory 48 GB VRAM/GPU OS Ubuntu 24.04 LTS Engine Ollama Backend ROCm Harness BoJl unified evaluation harness (2026-08) Prompt / output ? / 340 tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View Reportllama-bench	Devstral-2-123B-Instruct-2512Q3_K_SModel ref. ↗	Token speed512 input tokens · 0 existing KV tokens	—	104.9 tok/sPrefill · mean · per request	—	2 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Linux (inferred; distribution unreported) (inferred) Engine llama.cpp Backend vulkan_radv Harness llama-bench Prompt / output 512 / ? tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View Reportllama-bench	Devstral-2-123B-Instruct-2512Q3_K_SModel ref. ↗	Token speed512 input tokens · 0 existing KV tokens	—	194.63 tok/sPrefill · mean · per request	—	2 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Fedora 42 — Linux 6.15.9-201.fc42.x86_64 (Sat Aug 2 11:37:34 UTC 2025) Engine llama.cpp Backend vulkan_radv Harness llama-bench Prompt / output 512 / ? tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View ReportCustom evaluation	Devstral-2-123B-Instruct-2512Q4_K_MModel ref. ↗Conflicting evidence	Token speedsynth-poincare-section-pytest	—	—	6.96 tok/sGeneration · mean · per request	2 × W7800Layer split · 48 GB VRAM/GPU Configuration Memory 48 GB VRAM/GPU OS Ubuntu 24.04 LTS Engine Ollama Backend ROCm Harness BoJl unified evaluation harness (2026-08) Prompt / output ? / 932 tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View Reportllama-bench	Devstral-2-123B-Instruct-2512Q3_K_SModel ref. ↗	Token speed128 output tokens · 0 existing KV tokens	—	—	8.35 tok/sGeneration · mean · per request	2 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Linux (inferred; distribution unreported) (inferred) Engine llama.cpp Backend vulkan_radv Harness llama-bench Prompt / output ? / 128 tokens Context Not reported Concurrency Not reported Date Not reported Full record →
View Reportllama-bench	Devstral-2-123B-Instruct-2512Q3_K_SModel ref. ↗	Token speed128 output tokens · 0 existing KV tokens	—	—	8.29 tok/sGeneration · mean · per request	2 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Fedora 42 — Linux 6.15.9-201.fc42.x86_64 (Sat Aug 2 11:37:34 UTC 2025) Engine llama.cpp Backend vulkan_radv Harness llama-bench Prompt / output ? / 128 tokens Context Not reported Concurrency Not reported Date Not reported Full record →

Quantizations and software

Quantizations: Q3_K_S, Q4_K_M
Engines: llama.cpp, Ollama

Explore all 115 results →

This is a selection across sources, not a ranking or a matched experiment. Reading benchmark reports

GPU benchmark pages: W7800 · R9700