Instructions to use avlp12/GLM-5.3-Flash-Alis-MLX-8bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use avlp12/GLM-5.3-Flash-Alis-MLX-8bit 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("avlp12/GLM-5.3-Flash-Alis-MLX-8bit")

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 avlp12/GLM-5.3-Flash-Alis-MLX-8bit with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "avlp12/GLM-5.3-Flash-Alis-MLX-8bit"

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": "avlp12/GLM-5.3-Flash-Alis-MLX-8bit"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

MLX LM

How to use avlp12/GLM-5.3-Flash-Alis-MLX-8bit with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "avlp12/GLM-5.3-Flash-Alis-MLX-8bit"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "avlp12/GLM-5.3-Flash-Alis-MLX-8bit"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "avlp12/GLM-5.3-Flash-Alis-MLX-8bit",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Hermes Agent

How to use avlp12/GLM-5.3-Flash-Alis-MLX-8bit 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 "avlp12/GLM-5.3-Flash-Alis-MLX-8bit"

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 avlp12/GLM-5.3-Flash-Alis-MLX-8bit

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use avlp12/GLM-5.3-Flash-Alis-MLX-8bit with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "avlp12/GLM-5.3-Flash-Alis-MLX-8bit"

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 "avlp12/GLM-5.3-Flash-Alis-MLX-8bit" \
  --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"

Fails to load with mlx-vlm 0.7.6

by finding1 - opened Oct 5

Discussion

finding1

Oct 5

INFO - Pre-loading language model: GLM-5.3-Flash-Alis-MLX-8bit
INFO - Loading model: GLM-5.3-Flash-Alis-MLX-8bit
WARNING - Checkpoint has no weights for module(s): vision_tower. Disabling them.
ERROR - Error loading model GLM-5.3-Flash-Alis-MLX-8bit: Received 857 parameters not in model: 
language_model.model.layers.0.self_attn.b_proj.biases,
language_model.model.layers.0.self_attn.b_proj.scales,
language_model.model.layers.0.self_attn.b_proj.weight,
language_model.model.layers.0.self_attn.conv1d.weight,
language_model.model.layers.0.self_attn.forget_gate.A_log,
language_model.model.layers.0.self_attn.forget_gate.dt_bias,
language_model.model.layers.0.self_attn.forget_gate.f_a_proj.biases,
language_model.model.layers.0.self_attn.forget_gate.f_a_proj.scales,
language_model.model.layers.0.self_attn.forget_gate.f_a_proj.weight,
language_model.model.layers.0.self_attn.forget_gate.f_b_proj.biases,
language_model.model.layers.0.self_attn.forget_gate.f_b_proj.scales,
language_model.model.layers.0.self_attn.forget_gate.f_b_proj.weight,
language_model.model.layers.0.self_attn.g_a_proj.biases,
language_model.model.layers.0.self_attn.g_a_proj.scales,
language_model.model.layers.0.self_attn.g_a_proj.weight,
[...]
[repeat for additional layer numbers]
[...]
vision_model.blocks.0.attn.k_norm.weight,
vision_model.blocks.0.attn.proj.bias,
vision_model.blocks.0.attn.proj.weight,
vision_model.blocks.0.attn.q_norm.weight,
vision_model.blocks.0.attn.qkv.bias,
vision_model.blocks.0.attn.qkv.weight,
vision_model.blocks.0.mlp.down_proj.bias,
vision_model.blocks.0.mlp.down_proj.weight,
vision_model.blocks.0.mlp.gate_proj.bias,
vision_model.blocks.0.mlp.gate_proj.weight,
vision_model.blocks.0.mlp.up_proj.bias,
vision_model.blocks.0.mlp.up_proj.weight,
vision_model.blocks.0.norm1.weight,
vision_model.blocks.0.norm2.weight,
[...]
[repeat for additional block numbers]
[...]
vision_model.downsample.bias,
vision_model.downsample.weight,
vision_model.merger.down_proj.weight,
vision_model.merger.gate_proj.weight,
vision_model.merger.post_projection_norm.bias,
vision_model.merger.post_projection_norm.weight,
vision_model.merger.proj.weight,
vision_model.merger.up_proj.weight,
vision_model.patch_embed.proj.bias,
vision_model.patch_embed.proj.weight,
vision_model.post_layernorm.weight.

avlp12

Owner 27 days ago

I am sorry for the late reply; I did not see this discussion until now. Thank you for the report and the full log.

I found two problems on my side:

Tensor layout. I converted this build for the glm5_next layout of mlx-vlm 0.7.0rc0. mlx-vlm 0.7.1 changed that layout (for example self_attn.forget_gate.* → self_attn.*, self_attn.conv1d → self_attn.qkv_conv.conv, vision_model.* → vision_tower.*), so newer versions reject the old keys. The 857 parameters in your log are 510 tensors from the 34 linear-attention layers and 347 vision tensors.
Incomplete upload. Some files were missing from this repository (six expert shards, the weight index and the tokenizer/processor files). I restored them on 2026-10-16.

I am fixing the layout now. I rewrote only the tensor names in the shard headers; the tensor data is unchanged. A structural check against the mlx-vlm 0.7.8 loader passes, but I have not yet tested a real load and generation. I will upload the fix after that test and post an update here.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment