GLM-5.3-Flash Q8_0 needs 373.58 GiB (Rough estimate)
Needs 373.58 GiB: GLM-5.3-Flash Q8_0 at 8,192 tokens is an estimate of 318.04 GiB–373.58 GiB. Checkpoint zai-org/GLM-5.3-Flash. File is 340,981,966,112 bytes. No device is selected, so this is demand only, not a fit. Rough estimate.
GLM-5.3 is a separate catalog model. GLM-5.3 VRAM. GLM-5.3-Flash on 2× DGX Spark. DGX Spark 128 GB unified. MiaAI-Lab TensorFold 0.6 TP2. LimeChain TensorFold C8 comparison.
Specific checkpoints with sourced architecture and weight files. GGUF Q8_0 (Unsloth): 318.04 GiB–373.58 GiB at 8,192 tokens, 1 sequence, FP16 KV, llama.cpp.
Publisher and provenance
- PublisherZhipu AI
- Hugging Face
zai-org/GLM-5.3-Flash - Revision
eb9eb208eb0d988989d07a6a12d0fdeb5f52574a - Total parameters321,323,031,390
- Active parametersUnknown / not independently sourced
- Layers45
- Query heads64
- KV heads64 (hybrid)
- Head dimensionderived
- Recorded context capability1,048,576
- Completenesspartial · partial
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
- 317.56 GiB (340,981,966,112 bytes)
- Raw KV cache
- 0.35 GiB (377,487,360 bytes)
- Engine extras / workspace
- 0.13 GiB–55.66 GiB
- Total estimated demand
- 318.04 GiB–373.58 GiB
Published artifacts
| Quantization | Selected files | Bytes | Engine | Evidence |
|---|---|---|---|---|
| FP8 safetensors fp8 | 328,326,771,576 (305.78 GiB / 328.33 GB) | vllm, sglang, unspecified | file-metadata | |
| GGUF Q8_0 (Unsloth) gguf-q8_0 | 340,981,966,112 (317.56 GiB / 340.98 GB) | llama.cpp, ollama, unspecified | file-metadata | |
| INT4 theoretical | No published file | 160,661,515,695 (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 |
|---|---|---|---|
| FP8 safetensors | 306.49 GiB–354.64 GiB | 306.75 GiB–355.00 GiB | 307.81 GiB–356.44 GiB |
| GGUF Q8_0 (Unsloth) | 317.78 GiB–373.20 GiB | 318.04 GiB–373.58 GiB | 319.10 GiB–375.07 GiB |
| INT4 theoretical Theoretical rows are not downloadable artifacts. | 149.84 GiB–193.41 GiB | 150.10 GiB–193.79 GiB | 151.16 GiB–195.29 GiB |
Raw KV contribution
Related pages
Sources
- GLM-5.3-Flash model card
- config.json — revision eb9eb208eb0d988989d07a6a12d0fdeb5f52574a
- safetensors index — metadata.total_size=328326771576
- Artifact repo unsloth/GLM-5.3-Flash-GGUF
Assumptions that affect this result
- Requested context is within the recorded native window of 1048576 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.
- Weight bytes come from 8 published safetensors shards totaling 340981966112 bytes.
- This checkpoint has a vision component. Text-only inference does not include vision encoder or projector bytes.
- Multi-token prediction / draft weights exist and are not included unless enabled.
Calculator
Loading catalog snapshot…
Search or pick a catalog checkpoint to see memory demand.