Llamacpp Quantizations of gemma-1.1-2b-it
Using llama.cpp release b2589 for quantization.
Original model: https://huggingface.co/google/gemma-1.1-2b-it
Download a file (not the whole branch) from below:
Filename | Quant type | File Size | Description |
---|---|---|---|
gemma-1.1-2b-it-Q8_0.gguf | Q8_0 | 2.66GB | Extremely high quality, generally unneeded but max available quant. |
gemma-1.1-2b-it-Q6_K.gguf | Q6_K | 2.06GB | Very high quality, near perfect, recommended. |
gemma-1.1-2b-it-Q5_K_M.gguf | Q5_K_M | 1.83GB | High quality, recommended. |
gemma-1.1-2b-it-Q5_K_S.gguf | Q5_K_S | 1.79GB | High quality, recommended. |
gemma-1.1-2b-it-Q5_0.gguf | Q5_0 | 1.79GB | High quality, older format, generally not recommended. |
gemma-1.1-2b-it-Q4_K_M.gguf | Q4_K_M | 1.63GB | Good quality, uses about 4.83 bits per weight, recommended. |
gemma-1.1-2b-it-Q4_K_S.gguf | Q4_K_S | 1.55GB | Slightly lower quality with small space savings. |
gemma-1.1-2b-it-IQ4_NL.gguf | IQ4_NL | 1.56GB | Decent quality, similar to Q4_K_S, new method of quanting, recommended. |
gemma-1.1-2b-it-IQ4_XS.gguf | IQ4_XS | 1.50GB | Decent quality, new method with similar performance to Q4. |
gemma-1.1-2b-it-Q4_0.gguf | Q4_0 | 1.55GB | Decent quality, older format, generally not recommended. |
gemma-1.1-2b-it-Q3_K_L.gguf | Q3_K_L | 1.46GB | Lower quality but usable, good for low RAM availability. |
gemma-1.1-2b-it-Q3_K_M.gguf | Q3_K_M | 1.38GB | Even lower quality. |
gemma-1.1-2b-it-IQ3_M.gguf | IQ3_M | 1.30GB | Medium-low quality, new method with decent performance. |
gemma-1.1-2b-it-IQ3_S.gguf | IQ3_S | 1.28GB | Lower quality, new method with decent performance, recommended over Q3 quants. |
gemma-1.1-2b-it-Q3_K_S.gguf | Q3_K_S | 1.28GB | Low quality, not recommended. |
gemma-1.1-2b-it-Q2_K.gguf | Q2_K | 1.15GB | Extremely low quality, not recommended. |
Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski
- Downloads last month
- 900
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.