GLM-4.7-Flash-Razor-17B-A3B-E32of64
GLM-4.7-Flash with half of its routed experts removed by RAZOR, a training-free expert pruning method. Every MoE layer keeps 32 of its original 64 routed experts. No gradient updates or recovery training were applied: the retained weights are the base model's own weights.
| Base | This model | |
|---|---|---|
| Routed experts per MoE layer | 64 | 32 |
| Total parameters | 30B | 17.0B (16.7B excluding the MTP module) |
| Active parameters per token | ~3B | ~3B |
| Active experts per token (top-k) | 4 | 4 (unchanged) |
| Shared experts | 1 | 1 (unchanged) |
| Decoder layers | 47 | 47 (unchanged) |
| Context length | 202,752 | 202,752 (unchanged) |
Pruning touches only the routed expert pool. Attention, the shared expert, the embedding, the LM head and the router's remaining rows are untouched, so the compute per token is essentially unchanged while expert storage halves.
A 25% variant is available as GLM-4.7-Flash-Razor-24B-A3B-E48of64.
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "Nickyang/GLM-4.7-Flash-Razor-17B-A3B-E32of64"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id, dtype="bfloat16", device_map="auto",
)
messages = [{"role": "user", "content": "Explain mixture-of-experts routing."}]
inputs = tokenizer.apply_chat_template(
messages, add_generation_prompt=True, return_tensors="pt",
).to(model.device)
print(tokenizer.decode(model.generate(inputs, max_new_tokens=256)[0]))
Requires a Transformers build that contains the native glm4_moe_lite
implementation. The checkpoint stores bfloat16 weights in five safetensors
shards.
How the experts were selected
RAZOR asks whether the surviving computation can replace an expert's function, rather than how often or how strongly the expert fires. For a token routed to the set $S$ with normalized weights $w_j$, let $c=\sum_{j\in S} w_j f_j$ be the routed mixture and $r_j=f_j-c$ the consensus residual of expert $j$. Deleting a selected expert $i$ makes the router promote its highest-ranked unselected expert $r$, whose pseudo-weight $w_r$ is its score divided by the original selected score sum. With $\lambda$ the routed-output scale, the exact local output change is
$$ \delta_i=\lambda\lVert c-\tilde c^{-i}\rVert_2 =\lambda,\frac{\lVert w_i r_i - w_r r_r\rVert_2}{1-w_i+w_r}. $$
Scores are aggregated by conditional root mean square over the calibration tokens routed to each expert, and the highest-scoring experts are retained per layer. See the paper for the derivation and the code for the implementation.
Reproducing this checkpoint
Calibration used 32,768-token rows from RazorCal, the 2,048-sample multi-domain corpus released with RAZOR.
pip install razor-moe
razor saliency --model <path-to-GLM-4.7-Flash> \
--data data/RazorCal.json --max-len 32768 --out out/sal
razor prune --model <path-to-GLM-4.7-Flash> --saliency out/sal \
--method razor --ratio 0.5 --out out/pruned
kept_expert_indices.json in this repository is the keep-set manifest this
checkpoint was built from, in the format razor verify expects:
razor verify --pruned . --model <path-to-GLM-4.7-Flash>
Selection depends on the calibration draw, so an independent run reproduces the procedure rather than this exact expert set.
Limitations
Expert pruning is lossy. Benchmark retention and predictive fidelity do not guarantee stable generation: the paper reports that responses change in diversity, formatting and termination behaviour even where task accuracy is largely preserved. Evaluate on your own workload before deploying.
The calibration corpus is multi-domain but finite, so behaviour on domains far from it is not characterised by the released measurements. This repository publishes the 50% expert-removal setting only; the paper also reports 25%.
License and attribution
This is a derivative of GLM-4.7-Flash, released by Z.ai under the MIT license, and is distributed under the same license. The retained weights are the base model's own weights; RAZOR contributes the expert selection, not new parameters.
The RAZOR code is Apache-2.0. RazorCal records remain subject to their applicable upstream terms; see data/LICENSE-DATA.
Citation
@misc{song2026razorpruningreplaceableexperts,
title={RAZOR: Pruning Replaceable Experts in LLMs},
author={Mingyang Song and Mao Zheng},
year={2026},
eprint={2609.30465},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2609.30465},
}