xLAM-8x22b-r-IMat-GGUF

Llama.cpp imatrix quantization of Salesforce/xLAM-8x22b-r

Original Model: Salesforce/xLAM-8x22b-r
Original dtype: BF16 (bfloat16)
Quantized by: llama.cpp b3649
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
xLAM-8x22b-r.Q8_0/*	Q8_0	149.43GB	✅ Available	⚪ Static	✂ Yes
xLAM-8x22b-r.Q6_K/*	Q6_K	115.54GB	✅ Available	⚪ Static	✂ Yes
xLAM-8x22b-r.Q4_K/*	Q4_K	85.60GB	✅ Available	🟢 IMatrix	✂ Yes
xLAM-8x22b-r.Q3_K/*	Q3_K	67.80GB	✅ Available	🟢 IMatrix	✂ Yes
xLAM-8x22b-r.Q2_K/*	Q2_K	52.11GB	✅ Available	🟢 IMatrix	✂ Yes

All Quants

Filename	Quant type	File Size	Status	Uses IMatrix	Is Split
xLAM-8x22b-r.BF16/*	BF16	281.27GB	✅ Available	⚪ Static	✂ Yes
xLAM-8x22b-r.FP16/*	F16	281.27GB	✅ Available	⚪ Static	✂ Yes
xLAM-8x22b-r.Q8_0/*	Q8_0	149.43GB	✅ Available	⚪ Static	✂ Yes
xLAM-8x22b-r.Q6_K/*	Q6_K	115.54GB	✅ Available	⚪ Static	✂ Yes
xLAM-8x22b-r.Q5_K/*	Q5_K	99.98GB	✅ Available	⚪ Static	✂ Yes
xLAM-8x22b-r.Q5_K_S/*	Q5_K_S	96.99GB	✅ Available	⚪ Static	✂ Yes
xLAM-8x22b-r.Q4_K/*	Q4_K	85.60GB	✅ Available	🟢 IMatrix	✂ Yes
xLAM-8x22b-r.Q4_K_S/*	Q4_K_S	80.49GB	✅ Available	🟢 IMatrix	✂ Yes
xLAM-8x22b-r.IQ4_NL/*	IQ4_NL	79.79GB	✅ Available	🟢 IMatrix	✂ Yes
xLAM-8x22b-r.IQ4_XS/*	IQ4_XS	75.49GB	✅ Available	🟢 IMatrix	✂ Yes
xLAM-8x22b-r.Q3_K/*	Q3_K	67.80GB	✅ Available	🟢 IMatrix	✂ Yes
xLAM-8x22b-r.Q3_K_L/*	Q3_K_L	72.59GB	✅ Available	🟢 IMatrix	✂ Yes
xLAM-8x22b-r.Q3_K_S/*	Q3_K_S	61.51GB	✅ Available	🟢 IMatrix	✂ Yes
xLAM-8x22b-r.IQ3_M/*	IQ3_M	64.50GB	✅ Available	🟢 IMatrix	✂ Yes
xLAM-8x22b-r.IQ3_S/*	IQ3_S	61.51GB	✅ Available	🟢 IMatrix	✂ Yes
xLAM-8x22b-r.IQ3_XS/*	IQ3_XS	58.24GB	✅ Available	🟢 IMatrix	✂ Yes
xLAM-8x22b-r.IQ3_XXS/*	IQ3_XXS	54.91GB	✅ Available	🟢 IMatrix	✂ Yes
xLAM-8x22b-r.Q2_K/*	Q2_K	52.11GB	✅ Available	🟢 IMatrix	✂ Yes
xLAM-8x22b-r.Q2_K_S/*	Q2_K_S	48.10GB	✅ Available	🟢 IMatrix	✂ Yes
xLAM-8x22b-r.IQ2_M/*	IQ2_M	46.72GB	✅ Available	🟢 IMatrix	✂ Yes
xLAM-8x22b-r.IQ2_S.gguf	IQ2_S	42.60GB	✅ Available	🟢 IMatrix	📦 No
xLAM-8x22b-r.IQ2_XS.gguf	IQ2_XS	42.01GB	✅ Available	🟢 IMatrix	📦 No
xLAM-8x22b-r.IQ2_XXS.gguf	IQ2_XXS	37.89GB	✅ Available	🟢 IMatrix	📦 No
xLAM-8x22b-r.IQ1_M.gguf	IQ1_M	32.74GB	✅ Available	🟢 IMatrix	📦 No
xLAM-8x22b-r.IQ1_S.gguf	IQ1_S	29.65GB	✅ 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/xLAM-8x22b-r-IMat-GGUF --include "xLAM-8x22b-r.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/xLAM-8x22b-r-IMat-GGUF --include "xLAM-8x22b-r.Q8_0/*" --local-dir ./
# see FAQ for merging GGUF's

---

Inference

Simple chat template

<s>[INST] {user_prompt}[/INST] {assistant_response}</s>[INST] {next_user_prompt}[/INST]

Chat template with system prompt

<s>[INST] {user_prompt}[/INST] {assistant_response}</s>[INST] {system_prompt}

{next_user_prompt}[/INST]

Llama.cpp

llama.cpp/main -m xLAM-8x22b-r.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: xLAM-8x22b-r.Q8_0)
3. Run gguf-split --merge xLAM-8x22b-r.Q8_0/xLAM-8x22b-r.Q8_0-00001-of-XXXXX.gguf xLAM-8x22b-r.Q8_0.gguf
   • Make sure to point gguf-split to the first chunk of the split.

---

Got a suggestion? Ping me @legraphista!