Update README.md
Browse files
README.md
CHANGED
|
@@ -14,4 +14,31 @@ tags:
|
|
| 14 |
|
| 15 |
> Llama 3.2 3B Instruct by Meta is a lightweight, instruction-tuned large language model with 3.21 billion parameters, designed for efficient text-only, multilingual dialogue applications such as agentic retrieval, summarization, and instruction following. It employs an optimized transformer architecture with supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF) to align with human-like helpfulness and safety. Boasting an impressive context length capacity of up to 128,000 tokens, it supports multiple languages including English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai.
|
| 16 |
|
| 17 |
-
> The model is tailored for on-device and edge deployment, enabling fast, private, and low-latency inference while maintaining high-quality performance on benchmarks, making it suitable for mobile and resource-constrained environments. Llama 3.2 3B Instruct excels in text generation, rewriting, and dialogue tasks, and represents a robust solution for building advanced AI assistants with strong multilingual and instruction-following capabilities.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
|
| 15 |
> Llama 3.2 3B Instruct by Meta is a lightweight, instruction-tuned large language model with 3.21 billion parameters, designed for efficient text-only, multilingual dialogue applications such as agentic retrieval, summarization, and instruction following. It employs an optimized transformer architecture with supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF) to align with human-like helpfulness and safety. Boasting an impressive context length capacity of up to 128,000 tokens, it supports multiple languages including English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai.
|
| 16 |
|
| 17 |
+
> The model is tailored for on-device and edge deployment, enabling fast, private, and low-latency inference while maintaining high-quality performance on benchmarks, making it suitable for mobile and resource-constrained environments. Llama 3.2 3B Instruct excels in text generation, rewriting, and dialogue tasks, and represents a robust solution for building advanced AI assistants with strong multilingual and instruction-following capabilities.
|
| 18 |
+
|
| 19 |
+
## Model Files
|
| 20 |
+
|
| 21 |
+
| Model File name | Size | QuantType |
|
| 22 |
+
|---|---|---|
|
| 23 |
+
| Llama-3.2-3B-Instruct.BF16.gguf | 6.43 GB | BF16 |
|
| 24 |
+
| Llama-3.2-3B-Instruct.F16.gguf | 6.43 GB | F16 |
|
| 25 |
+
| Llama-3.2-3B-Instruct.F32.gguf | 12.9 GB | F32 |
|
| 26 |
+
| Llama-3.2-3B-Instruct.Q2_K.gguf | 1.36 GB | Q2_K |
|
| 27 |
+
| Llama-3.2-3B-Instruct.Q3_K_L.gguf | 1.82 GB | Q3_K_L |
|
| 28 |
+
| Llama-3.2-3B-Instruct.Q3_K_M.gguf | 1.69 GB | Q3_K_M |
|
| 29 |
+
| Llama-3.2-3B-Instruct.Q3_K_S.gguf | 1.54 GB | Q3_K_S |
|
| 30 |
+
| Llama-3.2-3B-Instruct.Q4_K_M.gguf | 2.02 GB | Q4_K_M |
|
| 31 |
+
| Llama-3.2-3B-Instruct.Q4_K_S.gguf | 1.93 GB | Q4_K_S |
|
| 32 |
+
| Llama-3.2-3B-Instruct.Q5_K_M.gguf | 2.32 GB | Q5_K_M |
|
| 33 |
+
| Llama-3.2-3B-Instruct.Q5_K_S.gguf | 2.27 GB | Q5_K_S |
|
| 34 |
+
| Llama-3.2-3B-Instruct.Q6_K.gguf | 2.64 GB | Q6_K |
|
| 35 |
+
| Llama-3.2-3B-Instruct.Q8_0.gguf | 3.42 GB | Q8_0 |
|
| 36 |
+
|
| 37 |
+
## Quants Usage
|
| 38 |
+
|
| 39 |
+
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
|
| 40 |
+
|
| 41 |
+
Here is a handy graph by ikawrakow comparing some lower-quality quant
|
| 42 |
+
types (lower is better):
|
| 43 |
+
|
| 44 |
+
|