Instructions to use beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4 with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4 with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4"

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

MLX LM

How to use beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4 with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Hermes Agent

How to use beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4 with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4"

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4 with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4"

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4

This model beezu/GLM-4.5-Iceblink-v3-106B-A12B-mlx-nvfp4 was converted to MLX format from zerofata/GLM-4.5-Iceblink-v3-106B-A12B using mlx-lm version 0.31.2.

About this quant

Just a normal nvfp4. Before the latest version, this model used a bf16 lm_head layer, but that has been replaced with a full nvfp4 across the board.

There are issues with doing any sort of hybrid quantization using mxfp4/mxfp8/nvfp4, and the patches I previously put in were good enough to create a model, but not good enough to run them properly. Big thanks to @circadesu for bringing these issues to my attention.

For now, I've removed the bf16 head layer and fallen back to a "normal" quant for both mxfp4 and nvfp4. I'm also deleting the mxfp4-mxfp8 and nvfp4-mxfp8 quants of this model entirely. I want to revisit this, but it will take substantial work that I'm not ready to take on quite yet.

Original Model Card

Iceblink v3

Iceblink

Version 3 · GLM-4.5 Air

❊

Overview

Decided to try tuning Air again after I saw Axolotl make some improvements on their training implementation and now that I know a lot more about what I'm doing. And wow. I think this came out pretty good.

This model is a creative writing and RP model. Supports reasoning and no reasoning with the usual GLM Air templates. Although reasoning off is recommended generally.

❊

SillyTavern Settings

Recommended Roleplay Format

Actions: In plaintext

Dialogue: "In quotes"

Thoughts: *In asterisks*

Recommended Samplers

Temp: 0.8 - 0.9

MinP: 0.05

TopP: 0.95 - 1.00

Instruct

GLM4.5 (no thinking): SillyTavern Preset

❊

Quantizations

GGUF

iMatrix

❊

Creation Process

Creation Process: SFT > SFT

SFT on approx 15.3 million tokens (11.7 million trainable), SFW / NSFW RP, instruct & chat data.

Then I tried out an idea I saw from ConicCat and trained the model for 8 epochs on 96 short stories (150k tokens) from light novels and human authors the internet said were good. This seems to have had a surprisingly positive effect on the prose without hurting the intelligence too much.

I went back to my usual higher LR's for this model. It turns out the GLM chat template was more cursed than I originally gave it credit for while training. It was a skill issue all along, go figure.

Axolotl Config

SFT (4×H200)

base_model: zai-org/GLM-4.5-Air
eot_tokens:
  - "<|user|>"
  - "<|endoftext|>"
chat_template_jinja: ./glm_air.jinja
 
plugins:
  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
 
load_in_8bit: false
load_in_4bit: true
 
quantize_moe_experts: true  # important
 
datasets:
  - path: ./data/nothink_dataset.jsonl
    type: chat_template
  - path: ./data/think_dataset.jsonl
    type: chat_template
 
dataset_prepared_path: last_run_prepared
val_set_size: 0.01
output_dir: ./GLM-Air-v4-SFT-1
 
adapter: qlora
lora_model_dir:
 
sequence_len: 10756
sample_packing: true
 
lora_r: 128
lora_alpha: 16
peft_use_rslora: true
lora_dropout: 0
lora_target_modules:
  - q_proj
  - v_proj
  - k_proj
  - o_proj
 
lora_target_parameters:
  - mlp.experts.gate_up_proj
  - mlp.experts.down_proj
 
lora_mlp_kernel: false
lora_qkv_kernel: false
lora_o_kernel: false
 
gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 2
optimizer: adamw_torch_8bit
lr_scheduler: cosine
learning_rate: 1e-5
 
bf16: auto
tf32: false
 
resume_from_checkpoint:
logging_steps: 1
flash_attention: true
 
warmup_ratio: 0.1
evals_per_epoch: 3
saves_per_epoch: 3
 
fsdp_config:
  fsdp_version: 2
  offload_params: false
  cpu_ram_efficient_loading: false
  auto_wrap_policy: TRANSFORMER_BASED_WRAP
  transformer_layer_cls_to_wrap: Glm4MoeDecoderLayer
  state_dict_type: FULL_STATE_DICT
  sharding_strategy: FULL_SHARD
  reshard_after_forward: true
  activation_checkpointing: true
 
# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

Writing SFT (2×H200)

base_model: ApocalypseParty/GLM-Air-v4-SFT-1-merged
eot_tokens:
  - "<|user|>"
  - "<|endoftext|>"
chat_template_jinja: ./glm_air.jinja
 
plugins:
  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
 
load_in_8bit: false
load_in_4bit: true
 
quantize_moe_experts: true  # important
 
datasets:
  - path: ./data/dataset_writing.jsonl
    type: chat_template
 
dataset_prepared_path: last_run_prepared
output_dir: ./GLM-Air-v4-SFT-1-writing
 
wandb_project: GLM-Air-v4-SFT
wandb_name: GLM-Air-v4-SFT-1-writing
 
adapter: qlora
lora_model_dir:
 
sequence_len: 4096
sample_packing: true
 
lora_r: 16
lora_alpha: 32
lora_dropout: 0
lora_target_modules:
  - q_proj
  - v_proj
  - k_proj
  - o_proj
 
lora_target_parameters:
  - mlp.experts.gate_up_proj
  - mlp.experts.down_proj
 
lora_mlp_kernel: false
lora_qkv_kernel: false
lora_o_kernel: false
 
gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 8
optimizer: adamw_torch_8bit
lr_scheduler: cosine
learning_rate: 9e-6
 
bf16: auto
tf32: false
 
resume_from_checkpoint:
logging_steps: 1
flash_attention: true
 
warmup_ratio: 0.1
saves_per_epoch: 1
 
fsdp_config:
  fsdp_version: 2
  offload_params: false
  cpu_ram_efficient_loading: false
  auto_wrap_policy: TRANSFORMER_BASED_WRAP
  transformer_layer_cls_to_wrap: Glm4MoeDecoderLayer
  state_dict_type: FULL_STATE_DICT
  sharding_strategy: FULL_SHARD
  reshard_after_forward: true
  activation_checkpointing: true
 
# save_first_step: true  # uncomment this to validate checkpoint saving works with your config