Potentailly broken weights, see https://x.com/mattshumer_/status/1832424499054309804?s=46
Once resolved, this message will go away

Reflection-Llama-3.1-70B-IMat-GGUF

Llama.cpp imatrix quantization of mattshumer/Reflection-Llama-3.1-70B

Original Model: mattshumer/Reflection-Llama-3.1-70B
Original dtype: FP32 (float32)
Quantized by: llama.cpp b3671
IMatrix dataset: here

• Files
  • IMatrix
  • Common Quants
  • All Quants
• Downloading using huggingface-cli
• Inference
  • 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
Reflection-Llama-3.1-70B.Q8_0/*	Q8_0	74.98GB	✅ Available	⚪ Static	✂ Yes
Reflection-Llama-3.1-70B.Q6_K/*	Q6_K	57.89GB	✅ Available	⚪ Static	✂ Yes
Reflection-Llama-3.1-70B.Q4_K.gguf	Q4_K	42.52GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.Q3_K.gguf	Q3_K	34.27GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.Q2_K.gguf	Q2_K	26.38GB	✅ Available	🟢 IMatrix	📦 No

All Quants

Filename	Quant type	File Size	Status	Uses IMatrix	Is Split
Reflection-Llama-3.1-70B.BF16/*	BF16	141.12GB	✅ Available	⚪ Static	✂ Yes
Reflection-Llama-3.1-70B.FP16/*	F16	141.12GB	✅ Available	⚪ Static	✂ Yes
Reflection-Llama-3.1-70B.Q8_0/*	Q8_0	74.98GB	✅ Available	⚪ Static	✂ Yes
Reflection-Llama-3.1-70B.Q6_K/*	Q6_K	57.89GB	✅ Available	⚪ Static	✂ Yes
Reflection-Llama-3.1-70B.Q5_K/*	Q5_K	49.95GB	✅ Available	⚪ Static	✂ Yes
Reflection-Llama-3.1-70B.Q5_K_S/*	Q5_K_S	48.66GB	✅ Available	⚪ Static	✂ Yes
Reflection-Llama-3.1-70B.Q4_K.gguf	Q4_K	42.52GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.Q4_K_S.gguf	Q4_K_S	40.35GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.IQ4_NL.gguf	IQ4_NL	40.05GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.IQ4_XS.gguf	IQ4_XS	37.90GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.Q3_K.gguf	Q3_K	34.27GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.Q3_K_L.gguf	Q3_K_L	37.14GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.Q3_K_S.gguf	Q3_K_S	30.91GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.IQ3_M.gguf	IQ3_M	31.94GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.IQ3_S.gguf	IQ3_S	30.91GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.IQ3_XS.gguf	IQ3_XS	29.31GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.IQ3_XXS.gguf	IQ3_XXS	27.47GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.Q2_K.gguf	Q2_K	26.38GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.Q2_K_S.gguf	Q2_K_S	24.47GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.IQ2_M.gguf	IQ2_M	24.12GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.IQ2_S.gguf	IQ2_S	22.24GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.IQ2_XS.gguf	IQ2_XS	21.14GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.IQ2_XXS.gguf	IQ2_XXS	19.10GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.IQ1_M.gguf	IQ1_M	16.75GB	✅ Available	🟢 IMatrix	📦 No
Reflection-Llama-3.1-70B.IQ1_S.gguf	IQ1_S	15.34GB	✅ 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/Reflection-Llama-3.1-70B-IMat-GGUF --include "Reflection-Llama-3.1-70B.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/Reflection-Llama-3.1-70B-IMat-GGUF --include "Reflection-Llama-3.1-70B.Q8_0/*" --local-dir ./
# see FAQ for merging GGUF's

---

Inference

Make sure to set the system prompt:
You are a world-class AI system, capable of complex reasoning and reflection. Reason through the query inside <thinking> tags, and then provide your final response inside <output> tags. If you detect that you made a mistake in your reasoning at any point, correct yourself inside <reflection> tags.

Chat template with system prompt

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

You are a world-class AI system, capable of complex reasoning and reflection. Reason through the query inside <thinking> tags, and then provide your final response inside <output> tags. If you detect that you made a mistake in your reasoning at any point, correct yourself inside <reflection> tags.<|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 Reflection-Llama-3.1-70B.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: Reflection-Llama-3.1-70B.Q8_0)
3. Run gguf-split --merge Reflection-Llama-3.1-70B.Q8_0/Reflection-Llama-3.1-70B.Q8_0-00001-of-XXXXX.gguf Reflection-Llama-3.1-70B.Q8_0.gguf
   • Make sure to point gguf-split to the first chunk of the split.

---

Got a suggestion? Ping me @legraphista!