Llama-3.2-1B-Instruct-IMat-GGUF

Llama.cpp imatrix quantization of meta-llama/Llama-3.2-1B-Instruct

Original Model: meta-llama/Llama-3.2-1B-Instruct
Original dtype: BF16 (bfloat16)
Quantized by: llama.cpp b3825
IMatrix dataset: here

• Files
  • IMatrix
  • Common Quants
  • All Quants
• Downloading using huggingface-cli
• Inference
  • Simple chat template
  • Chat template with system prompt
  • Llama.cpp
• FAQ
  • Why is the IMatrix not applied everywhere?
  • How do I merge a split GGUF?

---

Files

IMatrix

Status: ✅ Available
Link: here

Common Quants

Filename	Quant type	File Size	Status	Uses IMatrix	Is Split
Llama-3.2-1B-Instruct.Q8_0.gguf	Q8_0	1.32GB	✅ Available	⚪ Static	📦 No
Llama-3.2-1B-Instruct.Q6_K.gguf	Q6_K	1.02GB	✅ Available	⚪ Static	📦 No
Llama-3.2-1B-Instruct.Q4_K.gguf	Q4_K	807.69MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.Q3_K.gguf	Q3_K	690.84MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.Q2_K.gguf	Q2_K	580.87MB	✅ Available	🟢 IMatrix	📦 No

All Quants

Filename	Quant type	File Size	Status	Uses IMatrix	Is Split
Llama-3.2-1B-Instruct.BF16.gguf	BF16	2.48GB	✅ Available	⚪ Static	📦 No
Llama-3.2-1B-Instruct.FP16.gguf	F16	2.48GB	✅ Available	⚪ Static	📦 No
Llama-3.2-1B-Instruct.Q8_0.gguf	Q8_0	1.32GB	✅ Available	⚪ Static	📦 No
Llama-3.2-1B-Instruct.Q6_K.gguf	Q6_K	1.02GB	✅ Available	⚪ Static	📦 No
Llama-3.2-1B-Instruct.Q5_K.gguf	Q5_K	911.50MB	✅ Available	⚪ Static	📦 No
Llama-3.2-1B-Instruct.Q5_K_S.gguf	Q5_K_S	892.56MB	✅ Available	⚪ Static	📦 No
Llama-3.2-1B-Instruct.Q4_K.gguf	Q4_K	807.69MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.Q4_K_S.gguf	Q4_K_S	775.65MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.IQ4_NL.gguf	IQ4_NL	773.03MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.IQ4_XS.gguf	IQ4_XS	743.14MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.Q3_K.gguf	Q3_K	690.84MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.Q3_K_L.gguf	Q3_K_L	732.52MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.Q3_K_S.gguf	Q3_K_S	641.69MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.IQ3_M.gguf	IQ3_M	657.29MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.IQ3_S.gguf	IQ3_S	643.92MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.IQ3_XS.gguf	IQ3_XS	621.11MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.IQ3_XXS.gguf	IQ3_XXS	562.11MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.Q2_K.gguf	Q2_K	580.87MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.Q2_K_S.gguf	Q2_K_S	554.66MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.IQ2_M.gguf	IQ2_M	515.45MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.IQ2_S.gguf	IQ2_S	488.71MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.IQ2_XS.gguf	IQ2_XS	475.87MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.IQ2_XXS.gguf	IQ2_XXS	447.03MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.IQ1_M.gguf	IQ1_M	413.61MB	✅ Available	🟢 IMatrix	📦 No
Llama-3.2-1B-Instruct.IQ1_S.gguf	IQ1_S	393.55MB	✅ Available	🟢 IMatrix	📦 No

Downloading using huggingface-cli

If you do not have hugginface-cli installed:

pip install -U "huggingface_hub[cli]"

Download the specific file you want:

huggingface-cli download legraphista/Llama-3.2-1B-Instruct-IMat-GGUF --include "Llama-3.2-1B-Instruct.Q8_0.gguf" --local-dir ./

If the model file is big, it has been split into multiple files. In order to download them all to a local folder, run:

huggingface-cli download legraphista/Llama-3.2-1B-Instruct-IMat-GGUF --include "Llama-3.2-1B-Instruct.Q8_0/*" --local-dir ./
# see FAQ for merging GGUF's

---

Inference

Simple chat template

<|begin_of_text|><|start_header_id|>system<|end_header_id|>

Cutting Knowledge Date: December 2023
Today Date: 26 Jul 2024

<|eot_id|><|start_header_id|>user<|end_header_id|>

{user_prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>

{assistant_response}<|eot_id|><|start_header_id|>user<|end_header_id|>

{next_user_prompt}<|eot_id|>

Chat template with system prompt

<|begin_of_text|><|start_header_id|>system<|end_header_id|>

Cutting Knowledge Date: December 2023
Today Date: 26 Jul 2024

{system_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|>

{user_prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>

{assistant_response}<|eot_id|><|start_header_id|>user<|end_header_id|>

{next_user_prompt}<|eot_id|>

Llama.cpp

llama.cpp/main -m Llama-3.2-1B-Instruct.Q8_0.gguf --color -i -p "prompt here (according to the chat template)"

---

FAQ

Why is the IMatrix not applied everywhere?

According to this investigation, it appears that lower quantizations are the only ones that benefit from the imatrix input (as per hellaswag results).

How do I merge a split GGUF?

1. Make sure you have gguf-split available
   • To get hold of gguf-split, navigate to https://github.com/ggerganov/llama.cpp/releases
   • Download the appropriate zip for your system from the latest release
   • Unzip the archive and you should be able to find gguf-split
2. Locate your GGUF chunks folder (ex: Llama-3.2-1B-Instruct.Q8_0)
3. Run gguf-split --merge Llama-3.2-1B-Instruct.Q8_0/Llama-3.2-1B-Instruct.Q8_0-00001-of-XXXXX.gguf Llama-3.2-1B-Instruct.Q8_0.gguf
   • Make sure to point gguf-split to the first chunk of the split.

---

Got a suggestion? Ping me @legraphista!