---
license: mit
base_model: deepseek-ai/DeepSeek-R1-Distill-Llama-70B
pipeline_tag: text-generation
library_name: transformers
tags:
- deepseek-r1
- llama
- 70b
- lora-merged
- bf16
- conditional-stable
---

# DeepSeek 70B Final 7,816 — Standalone Merged BF16

Standalone BF16 model produced by merging the Final 7,816-row LoRA adapter into
`deepseek-ai/DeepSeek-R1-Distill-Llama-70B`.

- Base model license: MIT
- Unique training rows: 7,816
- Held-out Eval rows: 200
- Adapter required after download: No
- Merge method: memory-safe safetensors shard-by-shard LoRA delta merge
- Release status: Conditional Stable
- Zero-defect claim: No

The repository includes `MERGE_REPORT.json`, the original configuration and
tokenizer, a shard index, checksums, and standalone Smoke3 evidence.