Skip to content
Other

Smolder-DeepSeek-R1-Distill-70B

a reasoning model built to prevent hallucinations

2 min readupdated

A 70.55B-parameter large language model by sanyamsood that pairs DeepSeek-R1's step-by-step reasoning with the Smolder Epistemic hallucination-suppression architecture. Designed to generate logically sound answers to complex questions, it refuses to answer rather than fabricate facts when data is missing. Based on deepseek-ai/DeepSeek-R1-Distill-Llama-70B (80 layers, hidden dimension 8192), it's built for developers and researchers who need high factual accuracy when working on complex topics.

Key features

  • Smolder Epistemic hallucination-suppression architecture combined with DeepSeek-R1's chain-of-thought reasoning.
  • Calibrated abstention: the model refrains from generating an answer when it isn't confident in the facts.
  • Strong resistance to misleading questions, scoring 80.0% accuracy on the TruthfulQA MC2 benchmark.
  • Open MIT license, letting you freely use and integrate the model weights into your projects.

What you can do

  • Integrate it via the transformers library into local assistants and analytical pipelines to answer complex scientific and technical questions.
  • Deploy it in workflows where minimizing fabricated facts is critical, relying on its built-in calibrated abstention mechanism.

Example

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "sanyamsood/Smolder-DeepSeek-R1-Distill-70B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16, device_map="auto")
inputs = tokenizer("Explain why quantum entanglement does not violate the principle of locality.", return_tensors="pt").to("cuda")
print(tokenizer.decode(model.generate(**inputs, max_new_tokens=512)[0], skip_special_tokens=True))
Read moreDetails
TruthfulQA MC2: 80.0% accuracy on tests with false premises.
OpenAI SimpleQA CGA: 50.0% correct answers among attempted responses (with calibrated abstention).
Architecture: 70.55B parameters, 80 layers, d=8192, primary language English, loads in bfloat16.
Discuss Telegram 0 138

Will it run?

Only with RAM offload

Even 4-bit is larger than your VRAM. Some layers can stay in system RAM (llama.cpp, ComfyUI), but it will be several times slower.

  • BF16weights 131 GB · ready files145 GB
  • FP8 / INT8recommendedweights 66 GB73 GB
  • 4-bitweights 37 GB41 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: Won't fit any home GPU — renting is cheaper.

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

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    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.

    # sanyamsood/Smolder-DeepSeek-R1-Distill-70B
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download sanyamsood/Smolder-DeepSeek-R1-Distill-70B --local-dir ~/models/Smolder-DeepSeek-R1-Distill-70B
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

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.

Model family

New posts about versions of DeepSeek-R1-Distill-Llama-70B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 7All on Hugging Face
Merges 12All on Hugging Face

About the model

sanyamsood/Smolder-DeepSeek-R1-Distill-70B
Author
sanyamsood
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

1380

Ask AI about this model

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

Comments