GLM-4.5-Air VRAM & Memory Requirements
Specific checkpoints with sourced architecture and weight files. GGUF Q4_K_M (Unsloth): 69.53 GiB–71.36 GiB at 8,192 tokens, 1 sequence, FP16 KV, llama.cpp.
This result is an estimate, not a measured trace.Not verifiedFile-backed weights
Runs on
Publisher and provenance
- PublisherZhipu AI
- Hugging Face
zai-org/GLM-4.5-Air - Revision
a24ceef6ce4f3536971efe9b778bdaa1bab18daa - Total parameters110,468,824,832
- Active parametersUnknown / not independently sourced
- Layers46
- 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
- 67.96 GiB (72,975,748,384 bytes)
- Raw KV cache
- 1.44 GiB (1,543,503,872 bytes)
- Engine extras / workspace
- 0.13 GiB–1.96 GiB
- Total estimated demand
- 69.53 GiB–71.36 GiB
Published artifacts
| Quantization | Selected files | Bytes | Engine | Evidence |
|---|---|---|---|---|
| BF16 safetensors bf16 | 110,468,824,832 (102.88 GiB / 110.47 GB) | vllm, sglang, unspecified | file-metadata | |
| GGUF Q4_K_M (Unsloth) gguf-q4_k_m | Q4_K_M/GLM-4.5-Air-Q4_K_M-00001-of-00002.gguf, Q4_K_M/GLM-4.5-Air-Q4_K_M-00002-of-00002.gguf | 72,975,748,384 (67.96 GiB / 72.98 GB) | llama.cpp, ollama, unspecified | file-metadata |
| GGUF Q5_K_M (Unsloth) gguf-q5_k_m | Q5_K_M/GLM-4.5-Air-Q5_K_M-00001-of-00002.gguf, Q5_K_M/GLM-4.5-Air-Q5_K_M-00002-of-00002.gguf | 83,519,589,632 (77.78 GiB / 83.52 GB) | llama.cpp, ollama, unspecified | file-metadata |
| GGUF Q6_K (Unsloth) gguf-q6_k | Q6_K/GLM-4.5-Air-Q6_K-00001-of-00002.gguf, Q6_K/GLM-4.5-Air-Q6_K-00002-of-00002.gguf | 99,007,528,192 (92.21 GiB / 99.01 GB) | llama.cpp, ollama, unspecified | file-metadata |
| GGUF Q8_0 (Unsloth) gguf-q8_0 | Q8_0/GLM-4.5-Air-Q8_0-00001-of-00003.gguf, Q8_0/GLM-4.5-Air-Q8_0-00002-of-00003.gguf, Q8_0/GLM-4.5-Air-Q8_0-00003-of-00003.gguf | 117,456,506,240 (109.39 GiB / 117.46 GB) | llama.cpp, ollama, unspecified | file-metadata |
| INT4 theoretical | No published file | 55,234,412,416 (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 | 103.87 GiB–105.75 GiB | 104.94 GiB–106.86 GiB | 109.26 GiB–111.30 GiB |
| GGUF Q4_K_M (Unsloth) | 68.45 GiB–70.19 GiB | 69.53 GiB–71.36 GiB | 73.84 GiB–75.98 GiB |
| GGUF Q5_K_M (Unsloth) | 78.27 GiB–80.21 GiB | 79.35 GiB–81.38 GiB | 83.66 GiB–85.99 GiB |
| GGUF Q6_K (Unsloth) | 92.69 GiB–94.92 GiB | 93.77 GiB–96.09 GiB | 98.08 GiB–100.70 GiB |
| GGUF Q8_0 (Unsloth) | 109.87 GiB–112.44 GiB | 110.95 GiB–113.62 GiB | 115.26 GiB–118.23 GiB |
| INT4 theoretical Theoretical rows are not downloadable artifacts. | 51.93 GiB–67.25 GiB | 53.00 GiB–68.60 GiB | 57.32 GiB–73.91 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: 1,543,503,872 bytes (1.4375 GiB)
- 32,768 tokens: 6,174,015,488 bytes (5.7500 GiB)
Related pages
Sources
- GLM-4.5-Air model card
- config.json — revision a24ceef6ce4f3536971efe9b778bdaa1bab18daa
- safetensors index — metadata.total_size=110468824832
- Artifact repo unsloth/GLM-4.5-Air-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-Air-Q4_K_M-00001-of-00002.gguf, Q4_K_M/GLM-4.5-Air-Q4_K_M-00002-of-00002.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…
Search or pick a catalog checkpoint to see memory demand.
Search or pick a catalog checkpoint to see memory demand.