Skip to content
Language models

Blacksmith-1a

522M-параметровая LLM, обученная с нуля на iGPU

2 min readupdated

Blacksmith-1a от TriusAI — decoder-only transformer на 522M параметров (20 слоёв, d_model 1280, 10 голов, MHA без GQA), обученный с нуля, а не файнтюн готовой модели. Рассчитан на английский язык; авторы прямо называют его исследовательским артефактом, а не продуктом.

Ключевые особенности

  • Архитектура NanoGPT: 20 слоёв, FFN 5120, LayerNorm, RoPE с interleaved-парами, untied embeddings; это не transformers-архитектура, поэтому AutoModel не подойдёт.
  • Контекст обучали поэтапно: 4k → 8k → 16k → 32k → 64k; для выхода за trained length используется YaRN.
  • Есть GGUF для llama.cpp (архитектура gptneox): Q4_K_M на 327 MB и f16 на 1047 MB, с YaRN 2x и контекстом до 131072.
  • Честно заявленные ограничения: почти нет мировых знаний, нет арифметики, retrieval надёжен только примерно до 1k токенов, только английский.
  • В репозитории — код модели, обучения, оценки и конвертации, логи и шаблон чата; веса и код под MIT.

Пример

import torch
from code.model import NanoGPT
from code.tokenmaker import TokenizerVocabSize
from code.hyperparameters import d_model, n_heads, n_layers, max_seq_len
model = NanoGPT(TokenizerVocabSize, d_model, n_heads, n_layers, max_seq_len).cuda().to(torch.bfloat16)
ckpt = torch.load('blacksmith-1a.pt', map_location='cuda', weights_only=False)
model.load_state_dict(ckpt['model']); model.eval()
ПодробнееОбучение: претрейн ~9.9B токенов (SlimPajama 40%, Cosmopedia v2 35%, UltraChat 15%, OpenThoughts 10%), финальный loss 1.97; затем SFT и retrieval-SFT.
Железо: одна consumer iGPU Radeon 8060S с 96 GiB VRAM; GGUF Q4_K_M весит 327 MB, f16 — 1047 MB, контекст до 131072 через YaRN 2x.
Оценка: identity верно во всех формулировках; «столица Франции» — Paris в ~2–3 из 3 сэмплов; «2+2» — никогда.
Ограничения: только английский, нет надёжных мировых знаний и арифметики, retrieval надёжен лишь примерно до 1k токенов, нет tool use и safety-тюнинга.
Лицензия весов и кода — MIT; у обучающих корпусов свои условия, часть research-only/share-alike.
Discuss Telegram 0 61

Will it run?

Runs in GGUF F16

The most precise of the ready quantised files fits.

  • GGUF F16weights 1.0 GB · ready files1.7 GB
  • GGUF Q4_K_Mweights 0.3 GB · ready files1.0 GB

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: GGUF F16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

A ready script for the machine we advise above: paste it and the model downloads and starts.

  1. 1. Rent A30

    $0.26 an hour with the coupon. Pick the Ubuntu image when you create the machine.

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # TriusAI/blacksmith-1a — A30, GGUF F16
    # NVIDIA driver: CUDA 13 needs 580+. If older or missing, the current nvidia-open is installed (keep yours: KEEP_DRIVER=1)
    DRV=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader 2>/dev/null | head -1 | cut -d. -f1)
    case "$DRV" in ""|*[!0-9]*) DRV=0 ;; esac
    if [ "$DRV" -lt 580 ] && [ -z "${KEEP_DRIVER:-}" ]; then
      . /etc/os-release && \
      curl -fsSLO "https://developer.download.nvidia.com/compute/cuda/repos/ubuntu${VERSION_ID//./}/x86_64/cuda-keyring_1.1-1_all.deb" && \
      sudo dpkg -i cuda-keyring_1.1-1_all.deb && rm -f cuda-keyring_1.1-1_all.deb && \
      sudo apt-get update && \
      sudo DEBIAN_FRONTEND=noninteractive apt-get install -y nvidia-open || \
      { echo "Could not install the NVIDIA driver: see the error above. To keep the current one, run with KEEP_DRIVER=1."; exit 1; }
      echo "NVIDIA driver updated. Reboot the machine (sudo reboot) and run this same script again."
      exit 0
    fi
    curl -fsSL https://ollama.com/install.sh | sh
    ollama run hf.co/TriusAI/blacksmith-1a:F16
    Add more LoRAs to blacksmith-1a
  3. 3. Use it

    The chat opens right in the terminal. Leave with /bye, start again with the same ollama run command.

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

  • MTP (multi-token prediction)0.34.0from November 16, 2026new release

    1.5–2× generation speed

    If the model has MTP layers (DeepSeek-V3/R1, Qwen3-Next, GLM, MiMo and others), it predicts several tokens per step itself.

    How to enable: vLLM / SGLang: --speculative-config '{"method": "mtp", "num_speculative_tokens": 1}'. Check the model card.

    Earlier generations (2)
    • EAGLE-3 — Speculative decoding with a small draft head. For models without their own MTP layers.
    • EAGLE-2 / Medusa — Earlier draft-head generations. Still work if that is all you find.
  • FP8 KV cache0.34.0from November 16, 2026new release

    2× more context in the same memory

    Keeps the attention cache in 8 bits, so long contexts fit where they did not.

    How to enable: vLLM: --kv-cache-dtype fp8; llama.cpp: -ctk q8_0 -ctv q8_0.

  • vLLM / SGLang0.34.0from November 16, 2026new release

    many times more tokens/s on a server

    Server engines: batching, FlashAttention, FP8 and speculative decoding out of the box.

    How to enable: vllm serve <model> — an OpenAI-compatible API.

    Earlier generations (1)
    • llama.cpp — GGUF models on any GPU, Mac or even CPU; some layers can stay in RAM.

Model family

New posts about versions of blacksmith-1a (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

TriusAI/blacksmith-1a
Author
TriusAI
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

610

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments