Instructions to use RAS-LAB/GLM-4.7-Flash-24B-A3B-KREAP-Code with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use RAS-LAB/GLM-4.7-Flash-24B-A3B-KREAP-Code with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="RAS-LAB/GLM-4.7-Flash-24B-A3B-KREAP-Code")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("RAS-LAB/GLM-4.7-Flash-24B-A3B-KREAP-Code")
model = AutoModelForCausalLM.from_pretrained("RAS-LAB/GLM-4.7-Flash-24B-A3B-KREAP-Code", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use RAS-LAB/GLM-4.7-Flash-24B-A3B-KREAP-Code with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "RAS-LAB/GLM-4.7-Flash-24B-A3B-KREAP-Code"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "RAS-LAB/GLM-4.7-Flash-24B-A3B-KREAP-Code",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/RAS-LAB/GLM-4.7-Flash-24B-A3B-KREAP-Code

SGLang

How to use RAS-LAB/GLM-4.7-Flash-24B-A3B-KREAP-Code with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "RAS-LAB/GLM-4.7-Flash-24B-A3B-KREAP-Code" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "RAS-LAB/GLM-4.7-Flash-24B-A3B-KREAP-Code",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "RAS-LAB/GLM-4.7-Flash-24B-A3B-KREAP-Code" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "RAS-LAB/GLM-4.7-Flash-24B-A3B-KREAP-Code",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use RAS-LAB/GLM-4.7-Flash-24B-A3B-KREAP-Code with Docker Model Runner:
```
docker model run hf.co/RAS-LAB/GLM-4.7-Flash-24B-A3B-KREAP-Code
```

buugiiiiii commited on Sep 28

Commit

1330abb

verified ·

1 Parent(s): 33e1c73

Align the code-control card with the KREAP collection identity

Browse files

Update the public title, usage ID, and comparison labels while preserving the disclosure that this is a standard REAP control without Korean protection

Constraint: The collection name must not imply Korean hard protection
Confidence: high
Scope-risk: narrow
Tested: Editor title and status text inspected before commit
Not-tested: Inference behavior

Files changed (1) hide show

README.md +9 -9

README.md CHANGED Viewed

@@ -17,9 +17,9 @@ tags:
   - baseline
 ---
-# GLM-4.7-Flash-REAP-Code-48E
-Code-calibrated standard REAP one-shot MoE Expert pruning of [zai-org/GLM-4.7-Flash](https://huggingface.co/zai-org/GLM-4.7-Flash). This checkpoint intentionally does not use Korean hard protection.
 Routed Experts were pruned **64 → 48 per layer (25.0%)** while top-4 routing was kept unchanged.
 This percentage is the reduction in routed Expert count, not total model parameters.
@@ -29,7 +29,7 @@ is stored in BF16/F32 safetensors, and received **no fine-tuning**.
 | Status | Result |
 |---|---|
 | Checkpoint integrity checks | **PASS** |
-| Internal release gate | **NOT APPLICABLE AS A K-GUARD CANDIDATE** |
 | External reproduction | **NOT YET REPORTED** |
 | Deployment status | **EXPERIMENTAL COMPARISON CONTROL** |
@@ -39,7 +39,7 @@ model.
 ## What this checkpoint shows
-- Method: **REAP-Code baseline**
 - Korean MC macro: **53.5** (base 68.8, -15.3 pp)
 - Overall non-Korean composite: **60.4** (base 65.5, -5.1 pp)
 - Coding 2-benchmark macro: **74.0** (base 76.0, -2.0 pp)
@@ -49,7 +49,7 @@ model.
 ## Benchmark results
-| benchmark | base | K-Guard 12.5% | K-Guard force-fit 25% | K-Guard force-fit 50% | **REAP-Code 25% (this checkpoint)** |
 |---|---:|---:|---:|---:|---:|
 | **Korean (held-out harness)** |  |  |  |  |  |
 | KMMLU | 49.4 | 48.8 | 42.4 | 23.6 | 35.0 |
@@ -169,18 +169,18 @@ These are **internal comparative evaluation results**, not standardized leaderbo
 Machine-readable results and the exact survivor manifest are included under `evaluation/`.
-**Internal classification:** This is a code-centric REAP comparison control. K-Guard release
 criteria do not apply because Korean hard protection was intentionally omitted.
 ### Interpretation boundaries
-- At the same 48-Expert size, K-Guard-48E has a 9.3 pp higher Korean MC macro, while this checkpoint
   has a 4.0 pp higher overall non-Korean composite, a 4.6 pp higher LiveCodeBench score, and a
   3.6 pp higher coding 3-benchmark macro.
 - This is not a controlled single-variable comparison: the two checkpoints differ in both
   calibration data and survivor-selection rules. It cannot isolate the causal effect of hard
   protection.
-- Machine-readable results include direct K-Guard-48E versus REAP-Code-48E McNemar tests for 11
   non-Korean binary/pass-fail tasks. Direct paired tests for the Korean benchmark set, aggregate
   bootstrap confidence intervals, multiple-comparison correction, and repeated calibration seeds
   are absent.
@@ -203,7 +203,7 @@ code from this repository.
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
-model_id = "buugiiiiii/GLM-4.7-Flash-REAP-Code-48E"
 tokenizer = AutoTokenizer.from_pretrained(
     model_id,
     trust_remote_code=False,

   - baseline
 ---
+# GLM-4.7-Flash-24B-A3B-KREAP-Code
+Code-calibrated standard REAP one-shot MoE Expert pruning of [zai-org/GLM-4.7-Flash](https://huggingface.co/zai-org/GLM-4.7-Flash), named KREAP-Code within the project collection but not a Korean-protected KREAP checkpoint
 Routed Experts were pruned **64 → 48 per layer (25.0%)** while top-4 routing was kept unchanged.
 This percentage is the reduction in routed Expert count, not total model parameters.
 | Status | Result |
 |---|---|
 | Checkpoint integrity checks | **PASS** |
+| Internal release gate | **NOT APPLICABLE AS A KOREAN-PROTECTED KREAP CANDIDATE** |
 | External reproduction | **NOT YET REPORTED** |
 | Deployment status | **EXPERIMENTAL COMPARISON CONTROL** |
 ## What this checkpoint shows
+- Method: **KREAP-Code baseline**
 - Korean MC macro: **53.5** (base 68.8, -15.3 pp)
 - Overall non-Korean composite: **60.4** (base 65.5, -5.1 pp)
 - Coding 2-benchmark macro: **74.0** (base 76.0, -2.0 pp)
 ## Benchmark results
+| benchmark | base | KREAP 12.5% | KREAP force-fit 25% | KREAP force-fit 50% | **REAP-Code 25% (this checkpoint)** |
 |---|---:|---:|---:|---:|---:|
 | **Korean (held-out harness)** |  |  |  |  |  |
 | KMMLU | 49.4 | 48.8 | 42.4 | 23.6 | 35.0 |
 Machine-readable results and the exact survivor manifest are included under `evaluation/`.
+**Internal classification:** This is a code-centric REAP comparison control. KREAP release
 criteria do not apply because Korean hard protection was intentionally omitted.
 ### Interpretation boundaries
+- At the same 48-Expert size, KREAP-48E has a 9.3 pp higher Korean MC macro, while this checkpoint
   has a 4.0 pp higher overall non-Korean composite, a 4.6 pp higher LiveCodeBench score, and a
   3.6 pp higher coding 3-benchmark macro.
 - This is not a controlled single-variable comparison: the two checkpoints differ in both
   calibration data and survivor-selection rules. It cannot isolate the causal effect of hard
   protection.
+- Machine-readable results include direct KREAP-48E versus REAP-Code-48E McNemar tests for 11
   non-Korean binary/pass-fail tasks. Direct paired tests for the Korean benchmark set, aggregate
   bootstrap confidence intervals, multiple-comparison correction, and repeated calibration seeds
   are absent.
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
+model_id = "RAS-LAB/GLM-4.7-Flash-24B-A3B-KREAP-Code"
 tokenizer = AutoTokenizer.from_pretrained(
     model_id,
     trust_remote_code=False,