GLM-4.5 VRAM & Memory Requirements

Specific checkpoints with sourced architecture and weight files. GGUF Q4_K_M (Unsloth): 204.59 GiB–209.20 GiB at 8,192 tokens, 1 sequence, FP16 KV, llama.cpp.

This result is an estimate, not a measured trace.Not verifiedFile-backed weights

Publisher and provenance

PublisherZhipu AI
Hugging Facezai-org/GLM-4.5
Revisioncbb2c7cfb52fa128a9660cb1a7a78e017899e115
Total parameters358,337,791,296
Active parametersUnknown / not independently sourced
Layers92
Query heads96
KV heads8 (gqa)
Head dimension128
Recorded context capability131,072
Completenesscomplete · full

Active parameters describe tokens' routed compute, not the full resident model weight footprint. Full-resident weight estimates use total parameters.

Memory layers in this estimate

Artifact bytes, raw KV, engine extras, and total demand are separate. This is not a runtime trace.

Weight artifact: 201.59 GiB (216,455,572,672 bytes)
Raw KV cache: 2.88 GiB (3,087,007,744 bytes)
Engine extras / workspace: 0.13 GiB–4.73 GiB
Total estimated demand: 204.59 GiB–209.20 GiB

Published artifacts

Quantization	Selected files	Bytes	Engine	Evidence
BF16 safetensors bf16 Calculator: BF16 safetensors	93 safetensors shards Total selected bytes: 358,337,791,296 (333.73 GiB / 358.34 GB) First shard: model-00001-of-00093.safetensors Last shard: model-00093-of-00093.safetensors View all 93 shard files `model-00001-of-00093.safetensors` `model-00002-of-00093.safetensors` `model-00003-of-00093.safetensors` `model-00004-of-00093.safetensors` `model-00005-of-00093.safetensors` `model-00006-of-00093.safetensors` `model-00007-of-00093.safetensors` `model-00008-of-00093.safetensors` `model-00009-of-00093.safetensors` `model-00010-of-00093.safetensors` `model-00011-of-00093.safetensors` `model-00012-of-00093.safetensors` `model-00013-of-00093.safetensors` `model-00014-of-00093.safetensors` `model-00015-of-00093.safetensors` `model-00016-of-00093.safetensors` `model-00017-of-00093.safetensors` `model-00018-of-00093.safetensors` `model-00019-of-00093.safetensors` `model-00020-of-00093.safetensors` `model-00021-of-00093.safetensors` `model-00022-of-00093.safetensors` `model-00023-of-00093.safetensors` `model-00024-of-00093.safetensors` `model-00025-of-00093.safetensors` `model-00026-of-00093.safetensors` `model-00027-of-00093.safetensors` `model-00028-of-00093.safetensors` `model-00029-of-00093.safetensors` `model-00030-of-00093.safetensors` `model-00031-of-00093.safetensors` `model-00032-of-00093.safetensors` `model-00033-of-00093.safetensors` `model-00034-of-00093.safetensors` `model-00035-of-00093.safetensors` `model-00036-of-00093.safetensors` `model-00037-of-00093.safetensors` `model-00038-of-00093.safetensors` `model-00039-of-00093.safetensors` `model-00040-of-00093.safetensors` `model-00041-of-00093.safetensors` `model-00042-of-00093.safetensors` `model-00043-of-00093.safetensors` `model-00044-of-00093.safetensors` `model-00045-of-00093.safetensors` `model-00046-of-00093.safetensors` `model-00047-of-00093.safetensors` `model-00048-of-00093.safetensors` `model-00049-of-00093.safetensors` `model-00050-of-00093.safetensors` `model-00051-of-00093.safetensors` `model-00052-of-00093.safetensors` `model-00053-of-00093.safetensors` `model-00054-of-00093.safetensors` `model-00055-of-00093.safetensors` `model-00056-of-00093.safetensors` `model-00057-of-00093.safetensors` `model-00058-of-00093.safetensors` `model-00059-of-00093.safetensors` `model-00060-of-00093.safetensors` `model-00061-of-00093.safetensors` `model-00062-of-00093.safetensors` `model-00063-of-00093.safetensors` `model-00064-of-00093.safetensors` `model-00065-of-00093.safetensors` `model-00066-of-00093.safetensors` `model-00067-of-00093.safetensors` `model-00068-of-00093.safetensors` `model-00069-of-00093.safetensors` `model-00070-of-00093.safetensors` `model-00071-of-00093.safetensors` `model-00072-of-00093.safetensors` `model-00073-of-00093.safetensors` `model-00074-of-00093.safetensors` `model-00075-of-00093.safetensors` `model-00076-of-00093.safetensors` `model-00077-of-00093.safetensors` `model-00078-of-00093.safetensors` `model-00079-of-00093.safetensors` `model-00080-of-00093.safetensors` `model-00081-of-00093.safetensors` `model-00082-of-00093.safetensors` `model-00083-of-00093.safetensors` `model-00084-of-00093.safetensors` `model-00085-of-00093.safetensors` `model-00086-of-00093.safetensors` `model-00087-of-00093.safetensors` `model-00088-of-00093.safetensors` `model-00089-of-00093.safetensors` `model-00090-of-00093.safetensors` `model-00091-of-00093.safetensors` `model-00092-of-00093.safetensors` `model-00093-of-00093.safetensors`	358,337,791,296 (333.73 GiB / 358.34 GB)	`vllm, sglang, unspecified`	`file-metadata`
GGUF Q4_K_M (Unsloth) gguf-q4_k_m Calculator: GGUF Q4_K_M (Unsloth)	5 safetensors shards Total selected bytes: 216,455,572,672 (201.59 GiB / 216.46 GB) First shard: Q4_K_M/GLM-4.5-Q4_K_M-00001-of-00005.gguf Last shard: Q4_K_M/GLM-4.5-Q4_K_M-00005-of-00005.gguf View all 5 shard files `Q4_K_M/GLM-4.5-Q4_K_M-00001-of-00005.gguf` `Q4_K_M/GLM-4.5-Q4_K_M-00002-of-00005.gguf` `Q4_K_M/GLM-4.5-Q4_K_M-00003-of-00005.gguf` `Q4_K_M/GLM-4.5-Q4_K_M-00004-of-00005.gguf` `Q4_K_M/GLM-4.5-Q4_K_M-00005-of-00005.gguf`	216,455,572,672 (201.59 GiB / 216.46 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
GGUF Q5_K_M (Unsloth) gguf-q5_k_m Calculator: GGUF Q5_K_M (Unsloth)	6 safetensors shards Total selected bytes: 254,246,825,344 (236.79 GiB / 254.25 GB) First shard: Q5_K_M/GLM-4.5-Q5_K_M-00001-of-00006.gguf Last shard: Q5_K_M/GLM-4.5-Q5_K_M-00006-of-00006.gguf View all 6 shard files `Q5_K_M/GLM-4.5-Q5_K_M-00001-of-00006.gguf` `Q5_K_M/GLM-4.5-Q5_K_M-00002-of-00006.gguf` `Q5_K_M/GLM-4.5-Q5_K_M-00003-of-00006.gguf` `Q5_K_M/GLM-4.5-Q5_K_M-00004-of-00006.gguf` `Q5_K_M/GLM-4.5-Q5_K_M-00005-of-00006.gguf` `Q5_K_M/GLM-4.5-Q5_K_M-00006-of-00006.gguf`	254,246,825,344 (236.79 GiB / 254.25 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
GGUF Q6_K (Unsloth) gguf-q6_k Calculator: GGUF Q6_K (Unsloth)	6 safetensors shards Total selected bytes: 294,371,691,840 (274.16 GiB / 294.37 GB) First shard: Q6_K/GLM-4.5-Q6_K-00001-of-00006.gguf Last shard: Q6_K/GLM-4.5-Q6_K-00006-of-00006.gguf View all 6 shard files `Q6_K/GLM-4.5-Q6_K-00001-of-00006.gguf` `Q6_K/GLM-4.5-Q6_K-00002-of-00006.gguf` `Q6_K/GLM-4.5-Q6_K-00003-of-00006.gguf` `Q6_K/GLM-4.5-Q6_K-00004-of-00006.gguf` `Q6_K/GLM-4.5-Q6_K-00005-of-00006.gguf` `Q6_K/GLM-4.5-Q6_K-00006-of-00006.gguf`	294,371,691,840 (274.16 GiB / 294.37 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
GGUF Q8_0 (Unsloth) gguf-q8_0 Calculator: GGUF Q8_0 (Unsloth)	8 safetensors shards Total selected bytes: 380,966,784,608 (354.80 GiB / 380.97 GB) First shard: Q8_0/GLM-4.5-Q8_0-00001-of-00008.gguf Last shard: Q8_0/GLM-4.5-Q8_0-00008-of-00008.gguf View all 8 shard files `Q8_0/GLM-4.5-Q8_0-00001-of-00008.gguf` `Q8_0/GLM-4.5-Q8_0-00002-of-00008.gguf` `Q8_0/GLM-4.5-Q8_0-00003-of-00008.gguf` `Q8_0/GLM-4.5-Q8_0-00004-of-00008.gguf` `Q8_0/GLM-4.5-Q8_0-00005-of-00008.gguf` `Q8_0/GLM-4.5-Q8_0-00006-of-00008.gguf` `Q8_0/GLM-4.5-Q8_0-00007-of-00008.gguf` `Q8_0/GLM-4.5-Q8_0-00008-of-00008.gguf`	380,966,784,608 (354.80 GiB / 380.97 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
INT4 theoretical	No published file	179,168,895,648 (theoretical)	—	theoretical

Memory by quantization and context

Concurrent sequences = 1. KV defaults to FP16. The range includes documented runtime extras for the listed engine. These figures use the same core as the interactive calculator.

Quantization	2,048 tokens	8,192 tokens	32,768 tokens
BF16 safetensors	335.07 GiB–336.97 GiB	337.23 GiB–339.19 GiB	345.85 GiB–348.07 GiB
GGUF Q4_K_M (Unsloth)	202.43 GiB–206.85 GiB	204.59 GiB–209.20 GiB	213.21 GiB–218.43 GiB
GGUF Q5_K_M (Unsloth)	237.63 GiB–242.75 GiB	239.79 GiB–245.10 GiB	248.41 GiB–254.33 GiB
GGUF Q6_K (Unsloth)	275.00 GiB–280.87 GiB	277.16 GiB–283.21 GiB	285.78 GiB–292.44 GiB
GGUF Q8_0 (Unsloth)	355.65 GiB–363.13 GiB	357.80 GiB–365.48 GiB	366.43 GiB–374.70 GiB
INT4 theoretical Theoretical rows are not downloadable artifacts.	167.71 GiB–216.34 GiB	169.86 GiB–219.03 GiB	178.49 GiB–229.65 GiB

Raw KV contribution

Standard GQA/MHA raw KV is 2 × layers × kvHeads × headDim × tokens × sequences × bytes per element. This page uses KV heads (8), not query heads (96). Using query heads would overstate this cache by 12×.

8,192 tokens, 1 sequence, FP16: 3,087,007,744 bytes (2.8750 GiB)
32,768 tokens: 12,348,030,976 bytes (11.5000 GiB)

Open the calculator with this model

Sources

GLM-4.5 model card
config.json — revision cbb2c7cfb52fa128a9660cb1a7a78e017899e115
safetensors index — metadata.total_size=358337791296
Artifact repo unsloth/GLM-4.5-GGUF

Assumptions that affect this result

Requested context is within the recorded native window of 131072 tokens.
Full-resident weight estimates use total parameters, including inactive MoE experts.
Activated parameter count is unknown / not independently sourced. Active parameters describe routed compute, not the full resident model weight footprint.
Weight bytes come from the selected weight files only, not from every file in the repository.
Download file size is not peak GPU memory. Runtime layout, allocator padding, and KV are extra.
Counted files: Q4_K_M/GLM-4.5-Q4_K_M-00001-of-00005.gguf, Q4_K_M/GLM-4.5-Q4_K_M-00002-of-00005.gguf, Q4_K_M/GLM-4.5-Q4_K_M-00003-of-00005.gguf, Q4_K_M/GLM-4.5-Q4_K_M-00004-of-00005.gguf, Q4_K_M/GLM-4.5-Q4_K_M-00005-of-00005.gguf.
Multi-token prediction / draft weights exist and are not included unless enabled.
Context budget is the number of cached tokens per sequence (prompt plus reserved generation).

Calculator

Loading catalog snapshot…

Choose a model first, then check a device.

Model

Show historical models

Quantization

Device type

Hardware

Workload

Context length primarily affects KV cache memory.

Advanced settings

Runtime

Engine and KV cache type for this estimate.

EngineKV cache typeRecurrent / conv state dtype Include theoretical estimates (expert) Allow a substitute if the filter has no match

Workload

Overhead scenario and the prefill workspace.

Overhead scenario Include heavier prefill workspace

Topology

How the model is split across devices. Memory is not added together.

DeploymentIntra-replica splitDevice count

Replica count is an advanced deployment option. Quantity is nodes or GPUs, not replicas.

Replica groupsDevices per replica

Budget

System reserve and safety margin. An empty number keeps the previous value.

System / display reserve

GiB (2^30 bytes)

Budget already excludes system reserveExtra safety margin

Custom bits per weightDisplay unit

Custom model

Import a config or type the architecture. Nothing is scanned until you import.

Import local config.json

Parsed in the browser. The file is not uploaded.

Public Hugging Face repoRevision

Only huggingface.co, config.json, no token. Private repos are rejected.

Total parametersActive parametersLayersQuery headsKV headsHead dimension

Search or pick a catalog checkpoint to see memory demand.

Search or pick a catalog checkpoint to see memory demand.