GLM-4.7-Flash-REAP-23B-A3B GGUF size and VRAM requirements
unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF is a large language model with 23.0 billion parameters, built on the deepseek2 architecture. It is released under the mit license and has been downloaded 29,844 times.
GLM-4.7-Flash-REAP-23B-A3B-GGUF is a Mixture-of-Experts model with 48 experts, of which 4 are active on each token. Routing only 4 of 48 experts makes it noticeably faster than a dense model of the same size — but every expert still has to be held in memory, so the numbers below are set by the full parameter count, not the active one. It is built from 47 transformer layers, a hidden size of 2,048, 20 attention heads. GLM-4.7-Flash-REAP-23B-A3B-GGUF uses Multi-head Latent Attention, caching a single compressed latent per layer instead of full keys and values. That keeps the KV cache small even at long context — the memory figures on this page already reflect it.
To run unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF locally at a 4,096-token context, its quantized versions need between 7.09 GB (TQ1_0, lowest quality) and 43.86 GB (BF16, highest quality) of memory, weights plus KV cache and a system margin included. Context length drives that memory directly: at 4,096 tokens the KV cache for GLM-4.7-Flash-REAP-23B-A3B-GGUF is about 0.21 GB, rising to roughly 10.22 GB at its full 202,752-token context. Shorter prompts free up memory for a higher-quality quantization.
For most users the best balance is IQ1_S, needing about 7.72 GB. That means unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF fits entirely in the VRAM of an 8 GB GPU or larger, running fully on the GPU.
Available GGUF quantizations for unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF include TQ1_0, IQ1_S, IQ1_M, IQ2_XXS, IQ2_M, Q2_K, Q2_K_L, Q2_K_XL, IQ3_XXS, Q3_K_S, Q3_K_M, Q3_K_XL, IQ4_XS, IQ4_NL, Q4_0, Q4_K_S, Q4_K_M, Q4_K_XL, Q4_1, Q5_K_S, Q5_K_M, Q5_K_XL, Q6_K, Q6_K_XL, Q8_0, Q8_K_XL, BF16. The model supports a native context length of up to 202,752 tokens; a longer context grows the KV cache and the memory needed.
GGUF file size and memory by quantization
Compare real GGUF weight sizes, estimated KV cache and total memory for Q4, Q5, Q8 and every quantization published in this repository.
| Quant. | Bits | Quality | Weights | KV | Total | Speed~ | Verdict |
|---|---|---|---|---|---|---|---|
| TQ1_0 | 2.27 | Very low | 6.09 GB | 0.21 GB | 7.09 GB | 65.7 t/s | Fits in VRAM |
| IQ1_S | 2.51 | Very low | 6.71 GB | 0.21 GB | 7.72 GB | 59.6 t/s | Fits in VRAM |
| IQ1_M | 2.64 | Low | 7.07 GB | 0.21 GB | 8.07 GB | 7.1 t/s | Offload |
| IQ2_XXS | 2.87 | Low | 7.67 GB | 0.21 GB | 8.68 GB | 6.5 t/s | Offload |
| IQ2_M | 2.98 | Low | 7.97 GB | 0.21 GB | 8.98 GB | 6.3 t/s | Offload |
| Q2_K | 3.05 | Low | 8.17 GB | 0.21 GB | 9.17 GB | 6.1 t/s | Offload |
| Q2_K_L | 3.08 | Low | 8.24 GB | 0.21 GB | 9.24 GB | 6.1 t/s | Offload |
| Q2_K_XL | 3.14 | Low | 8.39 GB | 0.21 GB | 9.4 GB | 6.0 t/s | Offload |
| IQ3_XXS | 3.49 | Fair | 9.35 GB | 0.21 GB | 10.35 GB | 5.3 t/s | Offload |
| Q3_K_S | 3.58 | Fair | 9.59 GB | 0.21 GB | 10.6 GB | 5.2 t/s | Offload |
| Q3_K_M | 3.92 | Fair | 10.5 GB | 0.21 GB | 11.51 GB | 4.8 t/s | Offload |
| Q3_K_XL | 3.99 | Fair | 10.67 GB | 0.21 GB | 11.68 GB | 4.7 t/s | Offload |
| IQ4_XS | 4.37 | Good | 11.71 GB | 0.21 GB | 12.72 GB | 4.3 t/s | Offload |
| IQ4_NL | 4.61 | Good | 12.34 GB | 0.21 GB | 13.35 GB | 4.1 t/s | Offload |
| Q4_0 | 4.62 | Good | 12.38 GB | 0.21 GB | 13.38 GB | 4.0 t/s | Offload |
| Q4_K_S | 4.64 | Good | 12.41 GB | 0.21 GB | 13.42 GB | 4.0 t/s | Offload |
| Q4_K_M | 4.91 | Good | 13.14 GB | 0.21 GB | 14.15 GB | 3.8 t/s | Offload |
| Q4_K_XL | 4.96 | Good | 13.27 GB | 0.21 GB | 14.27 GB | 3.8 t/s | Offload |
| Q4_1 | 5.09 | Very good | 13.62 GB | 0.21 GB | 14.63 GB | 3.7 t/s | Offload |
| Q5_K_S | 5.58 | Very good | 14.94 GB | 0.21 GB | 15.94 GB | 3.3 t/s | Offload |
| Q5_K_M | 5.73 | Very good | 15.35 GB | 0.21 GB | 16.36 GB | 3.3 t/s | Offload |
| Q5_K_XL | 5.82 | Very good | 15.59 GB | 0.21 GB | 16.59 GB | 3.2 t/s | Offload |
| Q6_K | 6.61 | Excellent | 17.69 GB | 0.21 GB | 18.69 GB | 2.8 t/s | Offload |
| Q6_K_XL | 7.04 | Excellent | 18.84 GB | 0.21 GB | 19.85 GB | 2.7 t/s | Offload |
| Q8_0 | 8.51 | Excellent | 22.78 GB | 0.21 GB | 23.78 GB | 2.2 t/s | Offload |
| Q8_K_XL | 9.58 | Excellent | 25.64 GB | 0.21 GB | 26.65 GB | — | Insufficient |
| BF16 | 16.01 | Excellent | 42.85 GB | 0.21 GB | 43.86 GB | — | Insufficient |
KV cache computed from the model's exact architecture. Speed is a rough estimate bounded by memory bandwidth.
Frequently asked questions
What kind of model is unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF?
unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF is a language model with 23.0 billion parameters, based on the deepseek2 architecture. It is released under the mit license and distributed as GGUF files for local inference.
Is unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF a Mixture-of-Experts (MoE) model?
Yes. unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF is a Mixture-of-Experts model with 48 experts, of which 4 are activated per token. That makes it faster than a dense model of the same size, but all 48 experts must be loaded into memory, so the VRAM/RAM it needs is driven by the total parameter count, not the active one.
How does unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF handle long context?
unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF uses Multi-head Latent Attention, which caches one compressed latent per layer instead of full keys and values. Its KV cache therefore grows much more slowly than a standard model of the same size, so long contexts add relatively little memory.
How much VRAM do you need to run unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF?
You need about 7.72 GB of VRAM to run unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF entirely on the GPU using the IQ1_S quantization (at a 4,096-token context). Smaller quantizations lower the requirement at the cost of quality.
Can I run unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF on an 8 GB GPU?
Yes. With 8 GB of VRAM you can run unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF fully on the GPU using IQ1_S (about 7.72 GB).
Can I run unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF on a 16 GB GPU?
Yes. With 16 GB of VRAM you can run unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF fully on the GPU using Q5_K_S (about 15.94 GB).
Can I run unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF on a 24 GB GPU?
Yes. With 24 GB of VRAM you can run unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF fully on the GPU using Q8_0 (about 23.78 GB).
What context length does unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF support?
unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF supports a native context length of up to 202,752 tokens. A longer context grows the KV cache, so it increases the memory needed to run the model.
What is the best quantization for unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF?
For unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF, a strong default is Q4_K_M, which needs about 14.15 GB and keeps most of the quality while roughly halving the memory versus 8-bit. With VRAM to spare, Q5_K_M or Q6_K add a little more quality; if you are tight on memory, a smaller quantization still runs. Pick the highest quantization that fits your VRAM.