DeepSeek-R1-Distill-Llama-70B
- In Input modalities: Text Out Output modalities: Text
- Reasoning
- Open weights
Prices
0 offerings · prices as listedNo priced offering is listed for this model yet.
Details
- Tool use
- —
- Structured output
- —
- Licence
- llama3.3 license
- Family
- deepseek-r-distill-llama
Not stated on DeepSeek’s pages: max input, reasoning efforts, reasoning can be off, max thinking budget, built-in tools, knowledge cutoff, aliases.
Run locally
Weights: deepseek-ai/DeepSeek-R1-Distill-Llama-70B- Parameters 70.6B
- Dense
- Native precision BF16
- Layers 80
- KV cache per token 320 KB
- Config positions 128K (config.json max_position_embeddings)
- API context 125K (maker’s API)
- Needs
- ≈ 39.2 GB (weights 34.9 + KV cache 2.5 + overhead 1.7)
- Usable
- 21.6 GB (90% of its memory)
- In system memory
- 17.9 GB · 39/80 layers on the GPU (--n-gpu-layers 39)
- Generation speed
- 2–3 tok/s
- First token
- about 2.28–3.42 s for a 2K prompt
| Precision | Weights | Needs at 8K | On your hardware | Longest context |
|---|---|---|---|---|
| BF16 (native)maker’s file | 131.4 GB | ≈ 140.5 | ✕Doesn’t fit | — |
| Q8_0 | 69.8 GB | ≈ 75.8 | ✕Doesn’t fit | — |
| FP8 | 65.7 GB | ≈ 71.5 | ✕Doesn’t fit | — |
| Q6_K | 54.0 GB | ≈ 59.2 | ✕Doesn’t fit | — |
| Q5_K_M | 46.7 GB | ≈ 51.6 | ✕Doesn’t fit | — |
| Q5_0 | 45.2 GB | ≈ 49.9 | ✕Doesn’t fit | — |
| Q4_K_M | 39.9 GB | ≈ 44.4 | ○Needs system memory 2–3 tok/s | — |
| Q4_0 | 37.0 GB | ≈ 41.3 | ○Needs system memory 2–3 tok/s | — |
| IQ4_XS | 34.9 GB | ≈ 39.2 | ○Needs system memory 2–3 tok/s | — |
| Q3_K_Mquality drops | 32.1 GB | ≈ 36.2 | ○Needs system memory 3–4 tok/s | — |
| IQ3_XXSquality drops | 25.1 GB | ≈ 28.9 | ○Needs system memory 5–7 tok/s | — |
| Q2_Kquality drops | 24.6 GB | ≈ 28.4 | ○Needs system memory 5–7 tok/s | — |
| IQ2_XXSquality drops | 16.9 GB | ≈ 20.3 | ●Runs well 24–36 tok/s | 12K |
Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.
Compared with other hardware
Each at the highest precision it runs. Click one for details.
Smallest that runs it (Q4_K_M, 8K): GPU NVIDIA H100 SXM; unified memory Apple M5 Max (40-core GPU) · 64 GB.
Run commands
The maker publishes no files at this precision. Choose its native precision or one of its own builds, or quantise it yourself with llama.cpp.
Memory to fine-tune
Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.
Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-06. How it’s estimated
Cited scores
From third-party leaderboards, not measured by us- Epoch AI · GPQA Diamond 55.7%
#117 of 168 listed models · ±2.97
- Epoch AI · MATH Level 5 89.9%
#14 of 64 listed models · ±0.68
- Epoch AI · OTIS Mock AIME 2024–2025 51.4%
#111 of 159 listed models · ±6.05
- Epoch AI, 'Capabilities & benchmarking'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/benchmarks'. Licensed CC BY 4.0. · License · as of Oct 18 – Nov 4, 2026
- changes we made
Changes
All changes →No changes recorded yet.
Sources
- Alibaba Cloud Model Studio: text generation models (opens in a new tab) docs.modelstudio.console.alibabacloud.com · checked Sep 29, 2026
- deepseek-r1-distill-llama-70b - Alibaba Cloud Model Studio (opens in a new tab) docs.modelstudio.console.alibabacloud.com · checked Sep 29, 2026
- 模型上下架與更新in Chinese (opens in a new tab) alibabacloud.com · checked Sep 29, 2026
- deepseek-ai/DeepSeek-R1-Distill-Llama-70B model card (opens in a new tab) huggingface.co · checked Nov 1, 2026
- deepseek-ai/DeepSeek-R1-Distill-Llama-70B model card (opens in a new tab) huggingface.co · checked Nov 1, 2026
- deepseek-ai/DeepSeek-R1-Distill-Llama-70B model card (opens in a new tab) huggingface.co · checked Oct 30, 2026
- deepseek-ai/DeepSeek-R1-Distill-Llama-70B config.json (opens in a new tab) huggingface.co · checked Oct 31, 2026
- deepseek-ai/DeepSeek-R1-Distill-Llama-70B files (opens in a new tab) huggingface.co · checked Nov 3, 2026
DeepSeek’s official policies: data use, privacy, terms →
Some values on this page are still waiting to be checked against official pages: a filled dot is verified, a hollow one is not yet (with or without a source on file). How we verify data
DeepSeek · China · deepseek.com
Related models
Snapshots, the same family and other models from this provider