Computed

KAT Coder V2.5 Dev on an A100 40GB (SXM)

Yes, quantized. Q8_0 is the largest that fits, leaving room for 32k tokens of context.

Change the numbers →KAT Coder V2.5 Dev on every card A100 40GB (SXM) across every model

Verdict

Q8_0

36.8 GiB usable of 40 GB

Weights (Q4_K_M)

19.5 GiB

64.6 GiB at FP16 · 34.7B params · 1 of 6 quant rows are measured files

KV cache

80.0 KiB per token at FP16

40 layers × 2 KV heads × 256 dimensions

Speed ceiling

420 tok/s

1555 GB/s ÷ bytes read per token

Every quant, against this card

Weight bytes are the size of the real published file wherever one exists — 1 of these6 rows are measured from the published checkpoint, the rest computed from the parameter count. Max context is what the KV cache can grow to in whatever memory the weights leave behind, capped at the 256k tokens this model was trained to address.

Bars are the weight bytes at each precision; the dashed line is the usable memory of an A100 40GB (SXM). Drawn from the computed byte counts, not sketched.

Precision	Bits/weight	Weights	Source	Fits	Max context	Ceiling
FP16 / BF16	16.00	64.6 GiB	measured file	−27.8 GiB	—	27.2 tok/s offloaded
Q8_0	8.50	34.3 GiB	nominal	yes	32,797	420 tok/s
Q6_K	6.56	26.5 GiB	nominal	yes	135,400	516 tok/s
Q5_K_M	5.67	22.9 GiB	nominal	yes	182,470	577 tok/s
Q4_K_M	4.83	19.5 GiB	nominal	yes	226,896	650 tok/s
Q3_K_M	3.91	15.8 GiB	nominal	yes	262,144 (model cap)	753 tok/s

What the context actually costs

The KV cache is 2 · layers · kv_heads · head_dim bytes per token per element — 2 · 40 · 2 · 256 · 2 B = 80.0 KiB. It depends on the key/value head count, not the attention-head count and not the parameter count. This model shares 2 KV heads across 16 query heads, which divides the cache by 8 against multi-head attention.

Context	KV cache (FP16)	KV cache (8-bit)	Plus Q4_K_M weights
4,096	320 MiB	160 MiB	19.8 GiB
8,192	640 MiB	320 MiB	20.1 GiB
32,768	2.5 GiB	1.3 GiB	22.0 GiB
131,072	10.0 GiB	5.0 GiB	29.5 GiB

The speed ceiling, and where it comes from

Generating one token reads every active weight from memory once. At Q8_0 that is 2.8 GiB — only 8% of the checkpoint, because this is a mixture of experts and each token is routed to 8 of 256, plus a pass over the KV cache. An A100 40GB (SXM) moves 1555 GB/s, so the arithmetic ceiling is 420 tok/s. Treat it as a bound, not an estimate: attention overhead, kernel launches and imperfect memory access keep real runtimes at roughly 60–80% of it, and nothing pushes past it.

Where these numbers come from

The model

Parameters: 34,660,610,688
Layers: 40
Attention / KV heads: 16 / 2
Head dimension: 256
Trained context: 262,144
Checkpoint as published: 64.6 GiB

Read from Kwaipilot/KAT-Coder-V2.5-Dev. The parameter count is the Hub's own total over the tensor shapes, not a figure taken from the model's name.

The accelerator

Memory: 40 GB HBM2
Bandwidth: 1555 GB/s
Assumed usable: 92% → 36.8 GiB

Capacity and bandwidth from the vendor's specification. The usable fraction is an assumption, not a spec: a driver context, compute workspace and any attached display come out of the same pool before a weight is loaded.

Questions this pairing answers

How much VRAM does KAT Coder V2.5 Dev need?

64.6 GiB for the weights at FP16 — 34.7B parameters at two bytes each — and 19.5 GiB at Q4_K_M. The KV cache is on top of that and is not a fixed number: this model spends 80.0 KiB per token of context, so 8,192 tokens costs a further 640 MiB. An A100 40GB (SXM) makes 36.8 GiB of its 40 GB available on the assumption below.

Can an A100 40GB (SXM) run KAT Coder V2.5 Dev?

Yes, quantized. Q8_0 is the largest that fits, leaving room for 32k tokens of context. That is the weights and the KV cache together, against 36.8 GiB of usable memory.

How fast will KAT Coder V2.5 Dev run on an A100 40GB (SXM)?

No faster than 420 tokens/second at Q8_0, and in practice below it. Decoding is memory-bound: every token reads the 8% of weights this MoE routes to 2.8 GiB of weights plus the cache, and this card moves 1555 GB/s. That division is the ceiling — no kernel, runtime or driver beats it, and a real runtime typically reaches 60–80% of it.

Why does the context length change how much memory KAT Coder V2.5 Dev needs?

Because the KV cache holds one key and one value vector per token, per layer, for the whole conversation, and it is allocated separately from the weights. This model has 40 layers and 2 key/value heads of 256 dimensions, shared across 16 query heads — grouped-query attention, which divides the cache by 8. That works out at 80.0 KiB per token. Parameter count tells you nothing about this number.

The same model on a different card

A different model on the same card

All 62 models on A100 40GB (SXM) →

Method and limits. Weight bytes are the byte size of the real published file wherever one exists, and the model's exact parameter count times the published llama.cpp bits-per-weight where it does not. That distinction is on every row above and it matters at both ends: a sub-1B model's Q4_K_M file runs a third larger than the nominal figure because k-quants keep its embedding tables at higher precision, and an already-4-bit release cannot be quantized upward at all. The KV cache is 2 · layers · kv_heads · head_dim · bytes per token, summed over layers with each sliding-window layer capped at its window. The speed figure is a roofline bound, not a benchmark: bandwidth divided by bytes read per token, which no runtime exceeds and every runtime falls short of. The usable fraction of card memory is an assumption: 92% here, which is what this lane assumes for a dedicated card — the other class assumes 75%, so it is not one number applied to every device. That is the only assumed input on this page; every other figure is computed from the model config and the card's published specification. Nothing on this page is written by a language model. Architecture from the model's published config, fetched 2026-08-07. This checkpoint also carries a vision tower; its parameters are inside the totals above because they load whether or not you send it an image.