Qwen3.5-122B-A10B-GGUF

#!/usr/bin/env bash

custom="
# 60 Repeating Layers [0-59]

## Gated Attention/Delta Net [Blended 0-59]
blk\..*\.attn_gate\.weight=q8_0
blk\..*\.attn_qkv\.weight=q8_0
blk\..*\.attn_output\.weight=q8_0
blk\..*\.attn_q\.weight=q8_0
blk\..*\.attn_k\.weight=q8_0
blk\..*\.attn_v\.weight=q8_0
blk\..*\.ssm_alpha\.weight=f32
blk\..*\.ssm_beta\.weight=f32
blk\..*\.ssm_out\.weight=q8_0

# Shared Expert Layers [0-59]
blk\..*\.ffn_down_shexp\.weight=q8_0
blk\..*\.ffn_(gate|up)_shexp\.weight=q8_0

# Routed Experts Layers [0-59]
blk\..*\.ffn_down_exps\.weight=iq5_ks
blk\..*\.ffn_(gate|up)_exps\.weight=iq5_ks

# Non-Repeating Layers
token_embd\.weight=q8_0
output\.weight=q8_0
"

custom=$(
  echo "$custom" | grep -v '^#' | \
  sed -Ez 's:\n+:,:g;s:,$::;s:^,::'
)

    #--dry-run \
numactl -N ${SOCKET} -m ${SOCKET} \
./build/bin/llama-quantize \
    --custom-q "$custom" \
    --imatrix /mnt/data/models/ubergarm/Qwen3.5-122B-A10B-GGUF/imatrix-Qwen3.5-122B-A10B-BF16.dat \
    /mnt/data/models/ubergarm/Qwen3.5-122B-A10B-GGUF/Qwen3.5-122B-A10B-BF16-00001-of-00005.gguf \
    /mnt/data/models/ubergarm/Qwen3.5-122B-A10B-GGUF/Qwen3.5-122B-A10B-smol-IQ5_KS.gguf \
    IQ5_KS \
    128

# Clone and checkout
$ git clone https://github.com/ikawrakow/ik_llama.cpp
$ cd ik_llama.cpp

# Build for hybrid CPU+CUDA
$ cmake -B build -DCMAKE_BUILD_TYPE=Release -DGGML_CUDA=ON
$ cmake --build build --config Release -j $(nproc)

# Download Desired Quants
$ pip install huggingface_hub
$ hf download --local-dir ./ --include=smol-IQ2_KS/*.gguf ubergarm/Qwen3.5-122B-A10B-GGUF

# Full GPU Offload
./build/bin/llama-server \
  --model "$model" \
  --alias Qwen3.5-122B-A10B \
  -c 262144 \
  -fa on \
  -ger \
  --merge-qkv \
  -sm graph \
  -ngl 99 \
  -ub 4096 -b 4096 \
  --parallel 1 \
  --threads 1 \
  --host 127.0.0.1 \
  --port 8080 \
  --jinja \
  --no-mmap

# Hybrid CPU+GPU Offload
echo TODO or see other recent modelcards for examples running Qwen3.5

# CPU-Only Inference
numactl -N "$SOCKET" -m "$SOCKET" \
./build/bin/llama-server \
    --model "$model"\
    --alias ubergarm/Qwen3.5-122B-A10B \
    --ctx-size 65536 \
    -ctk q8_0 -ctv q8_0 \
    --parallel 1 \
    --threads 96 \
    --threads-batch 128 \
    --numa numactl \
    --host 127.0.0.1 \
    --port 8080 \
    --no-mmap \
    --jinja