K2-V2-Instruct
Where it runs 8K context · ≥ 5 t/s floor
| Setup | Best plan | Memory | Generate |
|---|---|---|---|
| 8 GB GPU + 32 GB RAM | Does not fit | ||
| 12 GB GPU + 32 GB RAM | Does not fit | ||
| 16 GB GPU + 64 GB RAM | Q4_K_Mllama.cpp CUDAGPU + CPU layers · 26/80 layers on GPU | Fits | ~0.7–0.8 t/s Estimated extrapolated |
| 24 GB GPU + 64 GB RAM | Q4_K_Mllama.cpp CUDAGPU + CPU layers · 41/80 layers on GPU | Fits | ~0.9–1.1 t/s Estimated extrapolated |
| 32 GB GPU + 128 GB RAM | Q4_K_Mllama.cpp CUDAGPU + CPU layers · 56/80 layers on GPU | Fits | ~1.4–1.7 t/s Estimated extrapolated |
| 2× 24 GB GPUs + 128 GB RAM | Q4_K_Mllama.cpp CUDALayer split across GPUs | Fits | ~12–15 t/s Estimated extrapolated |
| 96 GB workstation GPU + 256 GB RAM | Q8_0llama.cpp CUDAFull GPU | Fits | ~12–19 t/s Estimated |
| Mac, 24 GB | Does not fit | ||
| Mac, 48 GB | Does not fit | ||
| Mac, 128 GB | Q4_K_Mllama.cpp MetalUnified memory | Fits | ~7.0–11 t/s Estimated |
| Mac, 512 GB | Q8_0llama.cpp MetalUnified memory | Fits | ~4.5–6.9 t/s Estimated |
| DGX Spark 128 GB | Q8_0llama.cpp CUDAUnified memory | Fits | ~2.7–2.9 t/s Estimated |
| Ryzen AI Max+ 395, 128 GB | Q8_0llama.cpp VulkanUnified memory | Fits | ~1.8–2.6 t/s Estimated |
| CPU only, 64 GB DDR5-5600 dual channel | Q4_K_Mllama.cpp CPUCPU only | Fits | ~0.4–0.5 t/s Estimated extrapolated |
Same plan engine as the planner, with its default reserves and RAM. How plans are chosen.
Files exact Hugging Face bytes
| Artifact | Size | Files | Revision | Source |
|---|---|---|---|---|
| Q8_0 cturan/K2-V2-Instruct-GGUF | 71.81 GiB | 1 | 6ba45d7868 | HF files |
| Q4_K_M cturan/K2-V2-Instruct-GGUF | 40.89 GiB | 1 | 6ba45d7868 | HF files |
| Native safetensors float32 IFM/K2-V2-Instruct | 270 GiB | 6 | 6ef3e31ccd | HF files |
Not found in the representative quant search (not proven unavailable): GGUF FP16, GGUF Q5_K_M, MLX 8-bit, MLX 4-bit, EXL3, AWQ, GPTQ. Complete selected file sets; projectors, MTP/draft and lookup files are listed separately and budgeted only by plans that use them. File bytes are not allocated memory.
Quality
| AA evaluated mode | Index | LiveCodeBench | Terminal-Bench 2.1 | GPQA | HLE | Source |
|---|---|---|---|---|---|---|
| K2-V2 (High) AA estimate | 9.9 | 35.0% | — | 68.1% | 10.5% | source |
| K2-V2 (Medium) AA estimate | 9.0 | 28.0% | — | 59.8% | 4.4% | source |
| K2-V2 (Low) AA estimate | 7.3 | 20.3% | — | 54.1% | 3.6% | source |
Arena text rating
Unknown
Publisher-reported benchmarks
None extracted from the model card.
AA and Arena describe the hosted model in the stated mode, not each quantized file; scores are never inherited by base, pruned, abliterated or quantized variants and never averaged together.
How to run it
No model-specific recipe in the research. The planner binds the backend’s documented command template (llama.cpp, Ollama, LM Studio, MLX, vLLM) to your plan: open a plan’s details.
Recipes are source examples (settings are the publisher’s or author’s, not measured optima) and were not executed by LocalFit. A downloadable file does not prove your backend build supports the architecture.
Reported speeds 0
No reported measurement yet. Submit one.