### Run GLM-4.7-Flash-Claude-Opus with llama.cpp

Source: https://context7.com/teichai/glm-4.7-flash-claude-opus-4.5-high-reasoning-distill-gguf/llms.txt

Guide to running the GLM-4.7-Flash-Claude-4.5-Opus model directly using llama.cpp. It provides examples for general use, tool-calling/coding tasks, and setting up a local server.

```bash
# Download a quantized model file
# Available: q3_k_m, q3_k_s, q4_k_m, q5_k_m, q6_k, q8_0, iq2_m, iq3_m, iq3_xs, iq4_nl, iq4_xs, bf16, f16

# Run with recommended parameters for general use
./llama-cli -m glm-4.7-flash-claude-4.5-opus.q4_k_m.gguf \
  --temp 1.0 \
  --top-p 0.95 \
  --min-p 0.01 \
  --repeat-penalty 1.0 \
  -p "Explain quantum entanglement in simple terms"

# Run with parameters optimized for tool-calling/coding tasks
./llama-cli -m glm-4.7-flash-claude-4.5-opus.q4_k_m.gguf \
  --temp 0.7 \
  --top-p 1.0 \
  --min-p 0.01 \
  -p "Write a Python class for a binary search tree"

# Run as a server
./llama-server -m glm-4.7-flash-claude-4.5-opus.q4_k_m.gguf \
  --host 0.0.0.0 \
  --port 8080
```

--------------------------------

### llama.cpp - Server Mode

Source: https://context7.com/teichai/glm-4.7-flash-claude-opus-4.5-high-reasoning-distill-gguf/llms.txt

Starts a local HTTP server using llama.cpp to serve the model, allowing API-based interaction.

```APIDOC
## llama-server

### Description
Starts a local HTTP server using llama.cpp to serve the model, enabling API-based interaction.

### Method
CLI Command

### Endpoint
N/A (Local server)

### Parameters
#### Path Parameters
- **-m** (string) - Required - Path to the GGUF model file (e.g., `glm-4.7-flash-claude-4.5-opus.q4_k_m.gguf`).

#### Query Parameters
- **--host** (string) - Optional - The host address to bind the server to. Defaults to `0.0.0.0`.
- **--port** (integer) - Optional - The port number for the server. Defaults to `8080`.

### Request Example
```bash
./llama-server -m glm-4.7-flash-claude-4.5-opus.q4_k_m.gguf \
  --host 0.0.0.0 \
  --port 8080
```

### Response
Server starts and listens for incoming HTTP requests on the specified host and port. Output is typically server status messages.
```

--------------------------------

### GLM-4.7 Chat Template Format and Python Example

Source: https://context7.com/teichai/glm-4.7-flash-claude-opus-4.5-high-reasoning-distill-gguf/llms.txt

Demonstrates the specific chat template format required for multi-turn conversations with the GLM-4.7 model. Includes a Python function to correctly format prompts for the model.

```text
[gMASK]<sop><|system|>
You are a helpful assistant specialized in programming.<|user|>
How do I reverse a linked list in Python?<|assistant|>
Here is how to reverse a linked list...
```

```python
# Python example for formatting prompts
def format_glm_prompt(system: str, user: str) -> str:
    return f"[gMASK]<sop><|system|>
{system}<|user|>
{user}<|assistant|>
"

prompt = format_glm_prompt(
    system="You are an expert programmer.",
    user="Write a function to check if a number is prime."
)
print(prompt)
```

--------------------------------

### Download Quantized GLM-4.7 Model using Hugging Face CLI

Source: https://context7.com/teichai/glm-4.7-flash-claude-opus-4.5-high-reasoning-distill-gguf/llms.txt

Shows how to download a specific quantization of the GLM-4.7-Flash-Claude-Opus model from Hugging Face using the command-line interface. This allows users to select the desired balance between model size and quality.

```bash
# Download specific quantization from HuggingFace
huggingface-cli download TeichAI/GLM-4.7-Flash-Claude-Opus-4.5-High-Reasoning-Distill-GGUF \
  glm-4.7-flash-claude-4.5-opus.q4_k_m.gguf \
  --local-dir ./models
```

--------------------------------

### llama.cpp - CLI Usage

Source: https://context7.com/teichai/glm-4.7-flash-claude-opus-4.5-high-reasoning-distill-gguf/llms.txt

Demonstrates how to run the GLM-4.7-Flash-Claude-4.5-Opus model using the llama.cpp command-line interface for different tasks.

```APIDOC
## llama-cli

### Description
Run the model directly using llama.cpp with specified sampling parameters.

### Method
CLI Command

### Endpoint
N/A (Local execution)

### Parameters
#### Path Parameters
- **-m** (string) - Required - Path to the GGUF model file (e.g., `glm-4.7-flash-claude-4.5-opus.q4_k_m.gguf`).

#### Query Parameters
- **--temp** (float) - Optional - Temperature for sampling. Controls randomness. Defaults vary.
- **--top-p** (float) - Optional - Nucleus sampling probability. Defaults vary.
- **--min-p** (float) - Optional - Minimum probability for sampling. Defaults vary.
- **--repeat-penalty** (float) - Optional - Penalty for repeating tokens. Defaults vary.
- **-p** (string) - Required - The input prompt for the model.

### Request Example (General Use)
```bash
./llama-cli -m glm-4.7-flash-claude-4.5-opus.q4_k_m.gguf \
  --temp 1.0 \
  --top-p 0.95 \
  --min-p 0.01 \
  --repeat-penalty 1.0 \
  -p "Explain quantum entanglement in simple terms"
```

### Request Example (Tool-calling/Coding Tasks)
```bash
./llama-cli -m glm-4.7-flash-claude-4.5-opus.q4_k_m.gguf \
  --temp 0.7 \
  --top-p 1.0 \
  --min-p 0.01 \
  -p "Write a Python class for a binary search tree"
```

### Response
Output is printed directly to the console.

#### Success Response (Console Output)
- Text generated by the model based on the prompt and parameters.

#### Response Example
```text
Quantum entanglement is a phenomenon where two or more quantum particles become linked in such a way that they share the same fate, regardless of the distance separating them. Measuring a property of one particle instantaneously influences the corresponding property of the other particle(s). This connection is often described as 'spooky action at a distance' by Einstein.
```
```

--------------------------------

### Quantization Options

Source: https://context7.com/teichai/glm-4.7-flash-claude-opus-4.5-high-reasoning-distill-gguf/llms.txt

Lists available quantization options for the GLM-4.7-Flash-Claude-4.5-Opus model, detailing their file names, quality, and use cases.

```APIDOC
## Quantization Options

### Description
Select the appropriate quantization level based on your hardware capabilities and desired balance between model quality and resource consumption.

### Table

| Quantization Level | File Name Pattern                                     | Description                                             |
|--------------------|-------------------------------------------------------|---------------------------------------------------------|
| `bf16`             | `glm-4.7-flash-claude-4.5-opus.bf16.gguf`             | Highest quality, requires bfloat16 hardware support.    |
| `f16`              | `glm-4.7-flash-claude-4.5-opus.f16.gguf`              | High quality, standard float16 precision.               |
| `q8_0`             | `glm-4.7-flash-claude-4.5-opus.q8_0.gguf`             | Near-lossless 8-bit quantization, good balance.         |
| `q6_k`             | `glm-4.7-flash-claude-4.5-opus.q6_k.gguf`             | High quality 6-bit quantization.                        |
| `q5_k_m`           | `glm-4.7-flash-claude-4.5-opus.q5_k_m.gguf`           | Balanced quality and file size, good all-around option. |
| `q4_k_m`           | `glm-4.7-flash-claude-4.5-opus.q4_k_m.gguf`           | Recommended default, good quality with smaller size.    |
| `q3_k_m`           | `glm-4.7-flash-claude-4.5-opus.q3_k_m.gguf`           | Smaller file size, acceptable quality for resource limits.|
| `iq4_nl`           | `glm-4.7-flash-claude-4.5-opus.iq4_nl.gguf`           | Importance-weighted 4-bit quantization, efficient.      |
| `iq3_m`            | `glm-4.7-flash-claude-4.5-opus.iq3_m.gguf`            | Importance-weighted 3-bit quantization, very small size.|
| `iq2_m`            | `glm-4.7-flash-claude-4.5-opus.iq2_m.gguf`            | Smallest size, lower quality, for minimal resource use. |

### Downloading a Quantized Model
```bash
huggingface-cli download TeichAI/GLM-4.7-Flash-Claude-Opus-4.5-High-Reasoning-Distill-GGUF \
  glm-4.7-flash-claude-4.5-opus.q4_k_m.gguf \
  --local-dir ./models
```
```

--------------------------------

### Run GLM-4.7-Flash-Claude-Opus with Ollama

Source: https://context7.com/teichai/glm-4.7-flash-claude-opus-4.5-high-reasoning-distill-gguf/llms.txt

Instructions for loading and running the GLM-4.7-Flash-Claude-4.5-Opus model using Ollama. This includes creating the model from a Modelfile, running it interactively, and making API calls with curl.

```bash
# Create the model from the Modelfile
ollama create glm-claude-opus -f Modelfile

# Run the model interactively
ollama run glm-claude-opus

# Run with a specific prompt
ollama run glm-claude-opus "Explain the concept of recursion in programming"

# API usage with curl
curl http://localhost:11434/api/generate -d '{
  "model": "glm-claude-opus",
  "prompt": "Write a Python function to calculate factorial",
  "stream": false
}'
```

--------------------------------

### Chat Template Format

Source: https://context7.com/teichai/glm-4.7-flash-claude-opus-4.5-high-reasoning-distill-gguf/llms.txt

Illustrates the specific chat template format required for multi-turn conversations with the GLM-4.7 model.

```APIDOC
## Chat Template Format

### Description
Use the GLM-4.7 chat template format for structuring multi-turn conversations to ensure the model understands the context and roles.

### Format
```text
[gMASK]<sop><|system|>
{system_message}<|user|>
{user_message}<|assistant|>

```

### Parameters
- **[gMASK]** - Special token indicating the start of a masked generation.
- **<sop>** - Special token indicating the start of a new prompt.
- **<|system|>** - Role token for the system message.
- **{system_message}** - The system prompt or instructions.
- **<|user|>** - Role token for the user message.
- **{user_message}** - The user's input or query.
- **<|assistant|>** - Role token indicating the start of the assistant's response.

### Python Example
```python
def format_glm_prompt(system: str, user: str) -> str:
    return f"[gMASK]<sop><|system|>\n{system}<|user|>\n{user}<|assistant|>\n"

system_prompt = "You are an expert programmer."
user_query = "Write a function to check if a number is prime."

formatted_prompt = format_glm_prompt(system_prompt, user_query)
print(formatted_prompt)
```

### Example Output
```text
[gMASK]<sop><|system|>
You are an expert programmer.<|user|>
Write a function to check if a number is prime.<|assistant|>

```
```

--------------------------------

### Ollama API - Generate

Source: https://context7.com/teichai/glm-4.7-flash-claude-opus-4.5-high-reasoning-distill-gguf/llms.txt

This endpoint generates a response from the model based on a given prompt. It supports streaming responses and can be used with various models available in Ollama.

```APIDOC
## POST /api/generate

### Description
Generates a text completion or response from the specified model based on the provided prompt.

### Method
POST

### Endpoint
`http://localhost:11434/api/generate`

### Parameters
#### Request Body
- **model** (string) - Required - The name of the model to use (e.g., `glm-claude-opus`).
- **prompt** (string) - Required - The input text prompt for the model.
- **stream** (boolean) - Optional - If true, the response will be streamed in chunks. Defaults to false.

### Request Example
```json
{
  "model": "glm-claude-opus",
  "prompt": "Write a Python function to calculate factorial",
  "stream": false
}
```

### Response
#### Success Response (200)
- **response** (string) - The generated text response from the model.
- **done** (boolean) - Indicates if the generation is complete.
- **model** (string) - The name of the model used.

#### Response Example
```json
{
  "model": "glm-claude-opus",
  "created_at": "2023-10-30T10:00:00.123Z",
  "response": "def factorial(n):\n    if n == 0:\n        return 1\n    else:\n        return n * factorial(n-1)",
  "done": true,
  "context": [1, 2, 3, ...],
  "total_duration": 1234567890,
  "load_duration": 12345678,
  "prompt_eval_count": 10,
  "prompt_eval_duration": 1234567,
  "eval_count": 50,
  "eval_duration": 12345678
}
```
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.