LocalFit
Submit a benchmark

SERA-14B Specialist

allenai/SERA-14B

Released2026-02-03Size15B tensors · denseRoleSpecialistLicenseapache-2.0Context40K config max

Where it runs 8K context · ≥ 5 t/s floor

SetupBest planMemoryGenerate
8 GB GPU + 32 GB RAMQ4_K_Mllama.cpp CUDAGPU + CPU layers · 26/40 layers on GPUFits~4.9–6.0 t/s Estimated extrapolated
12 GB GPU + 32 GB RAMQ4_K_Mllama.cpp CUDAFull GPUFits~28–30 t/s Estimated
16 GB GPU + 64 GB RAMQ5_K_Mllama.cpp CUDAFull GPUFits~51–62 t/s Estimated
24 GB GPU + 64 GB RAMQ8_0llama.cpp CUDAFull GPUFits~35–44 t/s Estimated
32 GB GPU + 128 GB RAMQ8_0llama.cpp CUDAFull GPUFits~55–88 t/s Estimated
2× 24 GB GPUs + 128 GB RAMQ8_0llama.cpp CUDAFull GPUFits~32–42 t/s Estimated
96 GB workstation GPU + 256 GB RAMQ8_0llama.cpp CUDAFull GPUFits~57–89 t/s Estimated
Mac, 24 GBQ8_0llama.cpp MetalUnified memoryFits~12–14 t/s Estimated
Mac, 48 GBQ8_0llama.cpp MetalUnified memoryFits~12–14 t/s Estimated
Mac, 128 GBQ8_0llama.cpp MetalUnified memoryFits~19–29 t/s Estimated
Mac, 512 GBQ8_0llama.cpp MetalUnified memoryFits~21–33 t/s Estimated
DGX Spark 128 GBQ8_0llama.cpp CUDAUnified memoryFits~13–14 t/s Estimated
Ryzen AI Max+ 395, 128 GBQ8_0llama.cpp VulkanUnified memoryFits~8.5–12 t/s Estimated
CPU only, 64 GB DDR5-5600 dual channelQ4_K_Mllama.cpp CPUCPU onlyFits~2.1–2.5 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

ArtifactSizeFilesRevisionSource
Q8_0
mradermacher/SERA-14B-GGUF
14.62 GiB1ea03e65d03HF files
Q5_K_M
mradermacher/SERA-14B-GGUF
9.79 GiB1ea03e65d03HF files
Q4_K_M
mradermacher/SERA-14B-GGUF
8.38 GiB1ea03e65d03HF files
MLX 4-bit 4-bit
hitoshura25/SERA-14B-mlx-4Bit
7.74 GiB238ffd0d73cHF files
EXL3 4-bit
UnstableLlama/SERA-14B-exl3
8.15 GiB2b514d38e93HF files
Native safetensors BF16
allenai/SERA-14B
27.51 GiB69b585b4126HF files

Not found in the representative quant search (not proven unavailable): GGUF FP16, MLX 8-bit, 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

No Artificial Analysis evaluation mapped to this checkpoint.

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

vLLM · unknown

publisher-documented, not executed source

vllm serve allenai/SERA-14B --port 8001 \
            --tensor-parallel-size 4 \
            --max-model-len 32768 \
            --trust-remote-code \
            --enable-auto-tool-choice \
            --tool-call-parser hermes \
            --enforce-eager \
            --seed 42 \
            --disable-cascade-attn

Prerequisites: See source setup/dependencies; these snippets were inspected, not executed.

  • CodingagentoptimizedforClaudeCodeharness; plain chat benchmarks not same workflow.
  • Source TP4/32K/FP8 KV configuration is not a requirement nor fit proof for single consumer GPU.

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.