granite-20b-code-instruct-IMat-GGUF

Llama.cpp imatrix quantization of ibm-granite/granite-20b-code-instruct

Original Model: ibm-granite/granite-20b-code-instruct
Original dtype: BF16 (bfloat16)
Quantized by: llama.cpp b3649
IMatrix dataset: here


Files

IMatrix

Status: ✅ Available
Link: here

Common Quants

Filename Quant type File Size Status Uses IMatrix Is Split
granite-20b-code-instruct.Q8_0.gguf Q8_0 21.48GB ✅ Available ⚪ Static 📦 No
granite-20b-code-instruct.Q6_K.gguf Q6_K 16.63GB ✅ Available ⚪ Static 📦 No
granite-20b-code-instruct.Q4_K.gguf Q4_K 12.82GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.Q3_K.gguf Q3_K 10.57GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.Q2_K.gguf Q2_K 7.93GB ✅ Available 🟢 IMatrix 📦 No

All Quants

Filename Quant type File Size Status Uses IMatrix Is Split
granite-20b-code-instruct.BF16.gguf BF16 40.24GB ✅ Available ⚪ Static 📦 No
granite-20b-code-instruct.FP16.gguf F16 40.24GB ✅ Available ⚪ Static 📦 No
granite-20b-code-instruct.Q8_0.gguf Q8_0 21.48GB ✅ Available ⚪ Static 📦 No
granite-20b-code-instruct.Q6_K.gguf Q6_K 16.63GB ✅ Available ⚪ Static 📦 No
granite-20b-code-instruct.Q5_K.gguf Q5_K 14.81GB ✅ Available ⚪ Static 📦 No
granite-20b-code-instruct.Q5_K_S.gguf Q5_K_S 14.02GB ✅ Available ⚪ Static 📦 No
granite-20b-code-instruct.Q4_K.gguf Q4_K 12.82GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.Q4_K_S.gguf Q4_K_S 11.67GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.IQ4_NL.gguf IQ4_NL 11.55GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.IQ4_XS.gguf IQ4_XS 10.94GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.Q3_K.gguf Q3_K 10.57GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.Q3_K_L.gguf Q3_K_L 11.74GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.Q3_K_S.gguf Q3_K_S 8.93GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.IQ3_M.gguf IQ3_M 9.59GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.IQ3_S.gguf IQ3_S 8.93GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.IQ3_XS.gguf IQ3_XS 8.66GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.IQ3_XXS.gguf IQ3_XXS 8.06GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.Q2_K.gguf Q2_K 7.93GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.Q2_K_S.gguf Q2_K_S 7.15GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.IQ2_M.gguf IQ2_M 7.05GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.IQ2_S.gguf IQ2_S 6.53GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.IQ2_XS.gguf IQ2_XS 6.16GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.IQ2_XXS.gguf IQ2_XXS 5.57GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.IQ1_M.gguf IQ1_M 4.91GB ✅ Available 🟢 IMatrix 📦 No
granite-20b-code-instruct.IQ1_S.gguf IQ1_S 4.52GB ✅ 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/granite-20b-code-instruct-IMat-GGUF --include "granite-20b-code-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/granite-20b-code-instruct-IMat-GGUF --include "granite-20b-code-instruct.Q8_0/*" --local-dir ./
# see FAQ for merging GGUF's

Inference

Simple chat template

Question:
{user_prompt}

Answer:
{assistant_response}

Question:
{next_user_prompt}

Chat template with system prompt

System:
{system_prompt}

Question:
{user_prompt}

Answer:
{assistant_response}

Question:
{next_user_prompt}

Llama.cpp

llama.cpp/main -m granite-20b-code-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
  2. Locate your GGUF chunks folder (ex: granite-20b-code-instruct.Q8_0)
  3. Run gguf-split --merge granite-20b-code-instruct.Q8_0/granite-20b-code-instruct.Q8_0-00001-of-XXXXX.gguf granite-20b-code-instruct.Q8_0.gguf
    • Make sure to point gguf-split to the first chunk of the split.

Got a suggestion? Ping me @legraphista!

Downloads last month
342
GGUF
Model size
20.1B params
Architecture
starcoder

1-bit

2-bit

3-bit

4-bit

5-bit

6-bit

8-bit

16-bit

Inference Examples
Inference API (serverless) has been turned off for this model.

Model tree for legraphista/granite-20b-code-instruct-IMat-GGUF

Datasets used to train legraphista/granite-20b-code-instruct-IMat-GGUF

Evaluation results