Project Maya

GLM-5.3-Flash (321 B), DeepSeek-V4.1-Flash-1017 (748 B) and MiniMax-M3.1-Flash (428 B) on your own GPU
private and fast · one NVIDIA GPU or up to 16 · AMD and Windows (experimental) · chat, pictures, OpenAI- and Anthropic-compatible API · several requests at once

Created and led by Peasant Smith (@mw00), co-maintained with @needmorevram

_{Project Maya is free. If it runs well on your PC, a coffee keeps the work on it going.}

Models this size usually need a datacenter. Maya runs three of them on the hardware you already have: GLM-5.3-Flash, DeepSeek-V4.1-Flash-1017 and MiniMax-M3.1-Flash. Each model has an engine built for it. It keeps the experts your conversation uses most on the GPU, the next ones in RAM and the rest on your NVMe SSD, and moves them as you chat. It measures your GPUs, CPU, RAM and SSD and tunes itself to them. Nothing leaves your machine.

• Fast: up to 149 tokens/s on two RTX 5090s and an RTX 4090, 124 on two H100s, 118 on four RTX 4090s, 81 on one RTX PRO 6000, 30+ on a single RTX 5090. Real numbers from real users.
• Close to the original: Maya's own quants of GLM-5.3-Flash keep 97.7-99.2% of the FP8 model's zero-shot accuracy; DeepSeek V4.1 · Maya-S picks the same next token as the original 92.6% of the time, MiniMax M3.1 · Maya-S 87.6% (the average over its categories).
• Plugs into your tools: OpenAI- and Anthropic-compatible API with tool calls, ready for coding agents.
• Several requests at once: up to 8 answered together with GLM-5.3-Flash, DeepSeek-V4.1-Flash and MiniMax-M3.1-Flash - for agents that run several sessions, or several people on one PC.
• Everything in the box: a chat and a live monitor in the browser, pictures, thinking levels, a one-command setup.

Quick start

git clone https://github.com/mw00/project-maya.git && cd project-maya
./setup.sh

The setup asks which model you want, downloads it, builds everything once and starts the model. Then open http://127.0.0.1:8080. Windows: START-MAYA.bat instead of ./setup.sh. More in Install.

Pick your model

Every Maya quant is named after its model, then its tier - XS, S, M or L, by how many bits its experts keep. The setup recommends one for your PC; to pick one yourself, give its setup id: ./setup.sh --model glm-maya-s24.

Model	Maya quant · setup id	Size	Made for
GLM-5.3-Flash zai-org · MIT · 321 B, 18 B active · up to 1M context	Maya-S (default) · glm-maya-s	96.5 GB	PCs with a smaller memory pool (RAM and VRAM together)
	Maya-S24 · glm-maya-s24	94.7 GB	cards of 24 GB or less: ~1.5 GB more of the card for experts, decode +14%
	Maya-M · glm-maya-m	116 GB	a bigger memory pool; calibrated toward tool calls and front-end code
	Maya-M-Derisked · glm-maya-m-derisked	116 GB	Maya-M with a weight edit by Blackfrost_AI that reduces blanket refusals; experimental
	Maya-L · glm-maya-l	156.3 GB	the biggest memory pools; the closest to the original
DeepSeek-V4.1-Flash-1017 DeepSeek · MIT · 748 B	Maya-S (default) · deepseek-maya-s	340 GB	any amount of memory: what VRAM and RAM don't hold is read from the SSD
	Maya-XS · deepseek-maya-xs	277 GB	the whole model in a 128 GB PC's memory (the Engram tables stay on the SSD)
MiniMax-M3.1-Flash MiniMax · MiniMax Community License · 428 B, 23 B active · up to 1M context	Maya-XS · minimax-maya-xs	93 GB	the whole model in a 128 GB PC's memory
	Maya-S · minimax-maya-s	122 GB	closer to the original

All of them run on NVIDIA from the V100 / RTX 20 on, and on AMD and Windows (experimental; see What you need). DeepSeek's sizes include its Engram tables. The setup also offers unsloth's GGUF and LibertAIDAI's EXL3 quants of DeepSeek, orcarouter's abliterated DeepSeek, Mia-AiLab's EXL3 quant of MiniMax, turboderp's EXL3 quants of GLM, and Qwen-3.8-Flash-Next (CYBER-FROST) as its Q5_K_M GGUF, text only (its settings). Each model's files, drafters, pictures and options: docs/MODELS.md.

Quality

Maya's quants are made from each model's released weights and measured against the original:

• GLM-5.3 · Maya quants keep 97.7-99.2% of the FP8 model's zero-shot accuracy and pick its next token 83-90% of the time (Maya-L: 99.2%, 90%). How they're made and measured: MAYA-S, MAYA-S24, MAYA-M, MAYA-L.
• DeepSeek V4.1 · Maya-S picks the same next token as the original 92.6% of the time.
• MiniMax M3.1 · Maya-S picks it 87.6% of the time (the average over its categories).

More on each model's card on Hugging Face.

How fast is it?

Real machines measured by our users, most with ./maya.sh --bench (decode: three greedy 256-token answers; prefill: an 8K-token prompt unless the row says otherwise). Share yours: --bench, then --report, in a GitHub issue.

GLM-5.3-Flash speed benchmarks

Machine	Model	Decode (writing the answer)	Prefill (reading your prompt)	By
2x RTX 5090 32 GB + RTX 4090 48 GB (PCIe 5.0 x8, 5.0 x8, 4.0 x4), Core i7-13700K, 188 GB RAM	Maya-S / Maya-M-Derisked (DFlash2), 32K context	148.6 / 123.4 tokens/s	2605 / 2341 tokens/s	@prismaillya17 (v1.1.26, #498)
2x H100 80 GB (PCIe 5.0), Xeon Platinum 8488C, 504 GB RAM	Maya-L / Maya-M-Derisked, 1M context	122.4 / 123.7 tokens/s	2099 / 1986 tokens/s	@Molyleaf (v1.1.11, #258, #259)
4x RTX 4090 24 GB (PCIe 4.0 x16), Xeon w7-3445, 62 GB RAM	Maya-S24	118 tokens/s	4129 tokens/s	@0xPreDa (v1.0.31, #89)
8x RTX 3090 24 GB (PCIe 4.0, six x16 and two x8), EPYC 7402P, 126 GB RAM	EXL3-4.05bpw (DFlash2), 1M context	117.4 tokens/s	1470 tokens/s	@MarkMakers (v1.1.15, #414)
4x RTX 3090 24 GB	Maya-S24 / Maya-M (DFlash2), 256K context	100.8 / 86.5 tokens/s	979 / 907 tokens/s	@Whamp (v1.1.19, #349)
2x A100 80 GB (PCIe 4.0 x16), EPYC 7742 (a VM, 25 threads), 161 GB RAM	Maya-M, 128K context	91.9 tokens/s	-	@driona-will-fluff (v1.1.19, #326)
RTX PRO 6000 Blackwell 96 GB, Ryzen 7 9800X3D, 61 GB RAM	Maya-M, 256K context	81.0 tokens/s	3102 tokens/s	@yundddd (v1.1.7, #172)
4x Tesla V100 32 GB (PCIe 3.0 x16), 2x Xeon E5-2630 v4, 126-504 GB RAM, SATA SSD	Maya-S / Maya-M / Maya-L, 32K context	69.3 / 65.1 / 42.1 tokens/s	622 / 630 / 558 tokens/s	@HansStrupat (v1.1.19, #327, #340)
4x Tesla V100-SXM2 32 GB (NVLink 2.0)	Maya-S, 32K context	68.4 tokens/s	549 tokens/s	@jeremiahlangner (v1.1.18, #333)
1x H100 80 GB (PCIe 5.0 x16), Xeon Platinum 8488C, 504 GB RAM	Maya-M-Derisked, 1M context	61.4 tokens/s	2014 tokens/s	@Molyleaf (v1.1.15, #260)
2x NVIDIA CMP 170HX 64 GB (PCIe Gen2 x4 each), 2x Xeon E5-2690 v4, 91 GB RAM	Maya-M, every expert in VRAM	61.3 tokens/s	514 tokens/s	@ZackO2o (v1.0.6, #22)
2x RTX A6000 48 GB (PCIe 4.0 x8), Core i9-13900K, 128 GB RAM, SATA SSD	Maya-S / Maya-M / Maya-L, 131K-token prompts	60.7 / 50.9 / 32.1 tokens/s	1726 / 1629 / 1298 tokens/s (131K)	@everson (v1.1.20, #371)
2x RX 7900 XTX 24 GB (PCIe 4.0 x4 each) + Radeon 8060S helper (Ryzen AI Max+ 395, 128 GB)	Maya-M, up to 512K context	56.0 tokens/s (50.1 after a 456K-token prompt)	907 tokens/s (114K)	@pepuscz (v1.1.25 + #454-#457, #458)
2x RTX 5880 Ada 45 GB (PCIe 4.0 x16), Xeon Gold 5420+, 393 GB RAM	Maya-L, 256K context	54.5 tokens/s	1401 tokens/s	@vjsyong (v1.1.4, #149)
2x NVIDIA CMP 170HX 40 GB (PCIe Gen2 x16 each), Xeon w3-2423, 128 GB DDR5	Maya-S, 128K context	54.3 tokens/s	831 tokens/s	@nvsofts (v1.1.15, #263)
2x RTX 3090 24 GB (PCIe 4.0 x16, NVLink, power-capped at 252 W), Threadripper PRO 3975WX, 512 GB RAM	Maya-M / Maya-L, 128K context	54.1 / 40.3 tokens/s	936 / 850 tokens/s	@ksim63-gif (v1.1.23, #404)
2x RTX 3090 24 GB (PCIe 4.0 x16, power-capped at 225 W), Threadripper 3970X, 128 GB RAM	Maya-S24	53.0 tokens/s	885 tokens/s	@briepala (v1.1.2, #135)
3x Tesla V100 32 GB (PCIe 3.0 x16), 2x Xeon E5-2696 v3, 128 GB DDR3L, Windows 10	Maya-S, 128K context	52.9 tokens/s	616 tokens/s	@noahark (v1.1.20, #346)
4x Tesla V100 32 GB (PCIe 3.0 x16), 24 CPU cores, a 100 GiB RAM limit	Maya-L, 32K context	48.3 tokens/s (after --calibrate)	499 tokens/s	@areslp (v1.1.3, #130)
2x Radeon AI PRO R9700 32 GB (gfx1201, PCIe 4.0 x16 each, 240 W), EPYC 7313, 256 GB RAM	Maya-M-Derisked, 256K context	44.4 tokens/s	703 tokens/s	@knoodrake (v1.1.28, #509)
4x RTX 3090 24 GB (PCIe 4.0 x8 behind one switch), Xeon Platinum 8260M, 376 GB RAM	Maya-L, 1M context	44.2 tokens/s (tuned)	698 tokens/s	@foureight84 (v1.1.17, #302)
2x Radeon AI PRO R9700 32 GB (gfx1201, PCIe 4.0 x8 each), Ryzen 5 3600, 64 GB RAM	Maya-S24 / Maya-M-Derisked, 256K context	44.1 / 24.9 tokens/s	704 / 617 tokens/s	@gegnep (v1.1.20, #395)
2x RTX 3090 24 GB (PCIe 4.0 x16, power-capped at 250 W), EPYC 7452, 256 GB RAM	Maya-M / Maya-L, 128K context	44.0 / 30.2 tokens/s	860 / 801 tokens/s	@orochifj (v1.1.4, #165)
2x RTX 4090 24 GB (PCIe 3.0 + 4.0 x16), 2x Xeon Silver 4309Y, 188 GB RAM	Maya-S24	40.5 tokens/s	1271 tokens/s	@fuad00 (v1.1.0, #122)
RTX PRO 6000 Blackwell 96 GB (PCIe 4.0 x16), Ryzen 5 5600, 92 GB DDR4	Maya-L, 1M context	39.6 tokens/s	2708 tokens/s (3246 warm)	@QueTensors (v1.1.28, #532)
4x Tesla T10 15 GB (PCIe 3.0 x16), Xeon E5-2680 v4, 63 GB RAM	Maya-S24, 100K context	38.4 tokens/s	393 tokens/s	@zeallag (v1.1.16, #291)
2x Tesla V100 32 GB (PCIe 3.0 x8 each), Ryzen 9 7945HX3D, 96 GB RAM	Maya-M / Maya-L	33.8 / 13.3 tokens/s	490 / 372 tokens/s	@nkhaooat (v1.1.11, #208)
RTX 3090 24 GB + RTX 3060 12 GB (PCIe 4.0 x16), EPYC 7702P (8-channel DDR4), 141 GB RAM (a VM)	Maya-M, 128K context	33.7 tokens/s	682 tokens/s	@erikash (v1.1.4, #157)
RTX 5090 32 GB, Ryzen 9 7950X, 96 GB DDR5	Maya-M	30+ tokens/s	~2000 tokens/s	@siganos (#43)
2x RTX 3090 24 GB (PCIe 4.0 x16, NVLink), Threadripper PRO 3955WX, 247 GB RAM	EXL3-3.05bpw, 128K context	29.2 tokens/s	-	@catonooka (v1.1.9, #187)
2x Tesla P40 24 GB (Pascal, PCIe 3.0 x16), EPYC 7302P (a VM), 184 GB RAM	Maya-S24 / Maya-L	28.0 / 16.9 tokens/s	202 / 194 tokens/s	@Jw88888 (v1.1.11 + #214, #212)
RX 7900 XTX 24 GB + Radeon 8060S helper (Ryzen AI Max+ 395, 128 GB)	Maya-S24, 114K-token prompts	27.5 tokens/s	836 tokens/s (114K)	@pepuscz (v1.1.17, #311)
1x RTX 3090 24 GB (PCIe 4.0 x16), EPYC 7702P (8-channel DDR4), 141 GB RAM (a VM)	Maya-M, 128K context	27.0 tokens/s	1069 tokens/s	@erikash (v1.1.4, #157)
RTX 4080 SUPER 32 GB (a 32 GB modded card), Threadripper 3970X, 157 GB RAM, SATA SSD	Maya-S24, 262K context	26.7 tokens/s	1552 tokens/s	@Micro-ATP (v1.1.11, #200)
3x Tesla V100 16 GB (PCIe 3.0 x16), 2x Xeon E5-2690 v4, 62 GB RAM	Maya-S, 128K context	26.6 tokens/s	497 tokens/s	@saatvik-viniak (v1.1.11, #211)
4x RTX 3080 20 GB, 64 GB DDR4	Maya-M, 512K context	26.3 tokens/s (a day of use)	-	@CoolJohn-lab (#104)
1x RTX 3090 24 GB (power-capped at 225 W), Threadripper 3970X, 128 GB RAM	Maya-S24	26.0 tokens/s	913 tokens/s	@briepala (v1.1.2, #135)
RTX 4090 24 GB (PCIe 4.0 x16), 2x Xeon Silver 4310, 251 GB RAM	Maya-S24, 32K context	25.7 tokens/s	1963 tokens/s	@oscarlius (v1.1.24, #420)
RTX PRO 4000 Blackwell 24 GB, Core i9-13900K, 128 GB DDR5	Maya-S24	24.7 tokens/s	1146 tokens/s	@morrisfamily (v1.1.7, #174)
RTX 5090 32 GB (PCIe 5.0 x16), Core i9-13900K, 64 GB RAM, Windows 11	Maya-S, 128K context	23.0 tokens/s	2173 tokens/s	@gianbing (v1.1.3, #139)
4x Tesla T4 16 GB, Xeon E5-2660 v2, 220 GB RAM	Maya-S	22.6 tokens/s	416 tokens/s	@ksanislo (v1.0.16, #36)
Radeon AI PRO R9700 32 GB (gfx1201), Core i5-14400F, 128 GB RAM	Maya-S	21.4 tokens/s	-	@potatopote-debug (v1.1.12, #213)
1x NVIDIA CMP 170HX 64 GB (PCIe Gen2 x16), Core i7-9700K, 32 GB RAM	Maya-S24, 32K context	20.6 tokens/s	797 tokens/s	@Kanorin-chan (v1.1.8, #281)
AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB, 96 GB of it for the GPU), Windows 11	Maya-S	18.3 tokens/s	320 tokens/s	@LIN-LAB-AI (v1.0.30, #102)
RTX 3090 Ti 24 GB (PCIe 3.0 x16, a VM), EPYC 7302P, 240 GB RAM	Maya-L, 32K context	18.2 tokens/s (tuned, #177)	791 tokens/s	@Jw88888 (v1.1.7, #177)
3x Tesla P100 16 GB (Pascal; PCIe 3.0 x8, 3.0 x8, 2.0 x4), Ryzen 5 2600, 64 GB RAM	Maya-S24, 64K context	17.9 tokens/s	105 tokens/s	@maxlawrie (v1.1.28 + #479, #504)
RTX 4090 24 GB, Ryzen 7 9700X, 64 GB RAM	Maya-S24, 32K context	16.6 tokens/s	1320 tokens/s	@MagikEh (#274)
NVIDIA DGX Spark (GB10, 121 GB unified memory)	Maya-S, 128K context	16.5 tokens/s	~730 tokens/s (6K prompt)	@ericoh88 (v1.1.15, #282)
RTX 4090 24 GB, Ryzen 9 9950X3D, 192 GB DDR5, Windows 11	Maya-L, 128K context	16.4 tokens/s	1371 tokens/s	@npc97 (v1.0.24, #63)
RTX 5090 Laptop 24 GB, Core Ultra 9 275HX, 64 GB RAM, Windows 11 (a laptop)	Maya-S24	14.5 tokens/s	994 tokens/s	@klvnblst (v1.0.30, #67)

DeepSeek-V4.1-Flash speed benchmarks

Machine	Model	Decode (writing the answer)	Prefill (reading your prompt)	By
10x RTX 3090 24 GB (tensor parallel 2, PCIe 4.0 x4 each, 200 W), Xeon (24 threads), 64 GB RAM	EXL3 2.9 bpw, 1M context	97.0 tokens/s	2390 tokens/s (128K)	@DoktorMincs (v1.1.28, #494)
RTX 6000D 85 GB (PCIe 5.0 x16), Threadripper PRO 9965WX, 251 GB RAM	EXL3 2.9 bpw, 512K context	89.3 tokens/s	1687 tokens/s (2836 at 33K)	@Ross-Yang-94 (v1.1.25, #423)
2x A100-SXM4 80 GB, Xeon Platinum 8360Y, 1 TB RAM	Maya-S, 256K context	83.6 tokens/s	1207 tokens/s	@SunYong0821 (v1.1.24, #445)
2x RTX PRO 5000 Blackwell 72 GB, Ryzen 9 9950X3D, 249 GB DDR5	Maya-S, 512K context	68-71 tokens/s (prose), up to 162 (code), tuned	~3700 tokens/s (42K)	@realsquall1980-art (v1.1.28, #424)
4x AMD Instinct MI100 32 GB (gfx908, PCIe 4.0 x16 each), EPYC 7402, 252 GB RAM	Maya-XS, 32K context	51.8 tokens/s	172 tokens/s	@nsaxploit (v1.1.22, #396)
4x RTX 3090 24 GB (PCIe 3.0, SM clocks capped at 1500 MHz), Threadripper 1950X, 126 GB RAM	Maya-S, 256K context	50-53 tokens/s (code), 40-43 (prose)	947 tokens/s (10K), 1954 (90K)	@mgabor3141 (v1.1.26, #448)
2x NVIDIA A40 48 GB (PCIe 4.0 x16), 2x Xeon Gold 6330, 256 GB RAM, Windows Server 2022	Maya-S, 256K context	48.9 tokens/s (tuned)	~660 tokens/s (3K)	@yjyz1011 (v1.1.28, #508)
RTX 4090 24 GB (PCIe 4.0 x8), Ryzen 9 9950X3D, 192 GB DDR5	Maya-XS, 64K context	44.1 tokens/s	539 tokens/s	@naeq (v1.1.27, #503)
4x Tesla V100 32 GB (PCIe 3.0 x16), 2x Xeon E5-2630 v4, 504 GB RAM	Maya-S / UD-Q4_K_XL / UD-Q8_K_XL, 64K context	41.4 / 31 / 26.7 tokens/s	556 / 176 / 178 tokens/s	@HansStrupat (v1.1.26-v1.1.27, #467, #497)
2x RX 7900 XTX 24 GB (tensor parallel 2, PCIe 4.0 x4 each) + Radeon 8060S helper (Ryzen AI Max+ 395, 128 GB), the model on a USB4 NVMe	Maya-S, 512K context	41.3 tokens/s (46.1 after a 130K-token prompt, cached)	1291 tokens/s (130K; 482 at 8K)	@pepuscz (v1.1.28 + #511-#519, #510)
3x Tesla V100 32 GB (PCIe 3.0 x16), 2x Xeon E5-2696 v3, 128 GB DDR3L, Windows 10	Maya-S, 64K context	23-26 tokens/s (long answers)	377 tokens/s	@noahark (v1.1.26, #450)
2x RTX 4090 24 GB (PCIe 3.0 + 4.0 x16), 2x Xeon Silver 4309Y, 188 GB RAM	Maya-XS, 64K context	21.5 tokens/s	573 tokens/s	@fuad00 (v1.1.22, #393)
RTX 4000 Ada 20 GB, Xeon w7-3455, 255 GB DDR5, Windows 11	Maya-XS, 256K context	21.1 tokens/s	518 tokens/s (20K)	@pietroMastro92 (v1.1.29, #534)

MiniMax-M3.1-Flash speed benchmarks

Machine	Model	Decode (writing the answer)	Prefill (reading your prompt)	By
RTX 3090 24 GB + 8x RTX 5060 Ti 16 GB (PCIe 3.0 x8 each), 183 GB RAM	EXL3 3.5 bpw, tensor parallel 4 + DSpark, 8K context	38.2 tokens/s (code), 32.4 (prose)	-	@needmorevram (#477)
2x RTX 3090 24 GB (power-capped at 230 W), EPYC 7543 (a VM, 32 vCPUs), 224 GB RAM	Maya-S, 200K context, DSpark	22.0 tokens/s (code), 18.4 (prose)	163 tokens/s (10K)	@noonghunna (v1.1.30, #549)

A token is about ¾ of a word. The speed holds as a conversation grows, and the first answers after a start are the slowest while the caches fill with what you use.

Using it

• In the browser: chat (kept in your browser, with search, pins and folders, per-chat instructions and thinking level, edit and regenerate, HTML previews), a live Monitor of the model and your hardware (down to what fills each GPU's memory), a context size and port you can change in Settings, and Models: every model set up on this machine, switching to another (Maya starts again with it) and each one's engine settings (context, GPUs, drafting, images, memory), saved into its maya-<model>.json.
• Saved: the sidebar's Saved line adds up what the model's own API (DeepSeek's, Z.ai's or MiniMax's) would have charged for the same tokens - the prompt Maya reuses from its cache at the API's cached price, DeepSeek's peak hours at its peak price - kept per model in token-savings.json; its prices can be changed in Settings › Savings.
• Your apps and coding agents: an OpenAI-compatible API at http://127.0.0.1:8080/v1, and Anthropic's at http://127.0.0.1:8080/v1/messages (Claude Code: ANTHROPIC_BASE_URL=http://127.0.0.1:8080). Tool calls are streamed as the model writes them, and a long prompt's reading progress too when a request asks for it ("return_progress": true). A web page served from another address can call it too once you allow its address ("cors_origins", docs/SETUP.md).
• Responses API: /v1/responses supports text/image conversations, streaming, function tool calls, schema-validated output, background tasks with cancellation/replay, and optional shared in-memory history (store:true, previous_response_id). Responses are not stored by default. Options and limits: docs/RESPONSES.md.
• Thinking: Off, Low, High (the default) or Max, in Settings or per request.
• Pictures: attach one in the chat, or send image parts through either API.
• From another device: ./maya.sh --setup --host 0.0.0.0 --api-key <secret>. Always set a key.
• Tuning: ./maya.sh --calibrate tunes the engine to your hardware in 10-15 minutes. Every setting: docs/SETTINGS.md.
• Each model's own options - GLM-5.3-Flash's several requests at once (--parallel), long contexts and DFlash2 drafter; DeepSeek's kept conversations, expert loading and Engram read-ahead; MiniMax's thinking levels: docs/MODELS.md.

What you need

GPU	NVIDIA, compute capability 7.0 or newer (V100, RTX 20 and newer): one, or up to 16 that share the model. More VRAM is faster. Pascal (Tesla P40 / P100, GTX 10) is experimental, not validated yet. AMD (experimental): RX 7900 XT / XTX, Radeon AI PRO R9700 / RX 9070, Strix Halo and Gorgon Halo (Radeon 8060S / 8065S); RX 6800 / 6900 and Radeon PRO V620 / W6800 not validated yet; GLM-5.3-Flash reads text only there, DeepSeek-V4.1-Flash-1017 pictures too - docs/AMD_MAYA.md
RAM	32 GB runs it. More RAM keeps more experts close and is faster
Disk	A fast NVMe SSD with room for the model (~100 GB for GLM-5.3 · Maya-S, ~160 GB for GLM-5.3 · Maya-L, ~350 GB for DeepSeek V4.1 · Maya-S, ~520 GB for unsloth's GGUF of it, ~95 GB for MiniMax M3.1 · Maya-XS, ~200 GB for MiniMax-M3.1-Flash's EXL3 quant) and its picture encoder (1.1 GB). Not a hard disk
System	Linux (x86-64; a CPU with AVX2 is best), the NVIDIA driver, CUDA 12.x (CUDA 13 works for RTX 20 and newer, not for the V100 or Pascal), g++, Python 3.10+. Windows 10/11 (experimental): docs/WINDOWS.md. Not WSL2. AMD: ROCm 7

The setup checks all of this, prints the exact command for anything missing, and installs nothing system-wide.

Install

Run the two commands in Quick start and answer a few questions (Enter takes the recommended choice). The setup shows every download before it starts it, builds everything once (20-40 minutes plus the download) and starts the model. Then open http://127.0.0.1:8080. Ctrl+C stops it, and ./setup.sh starts it again right away.

• Windows: START-MAYA.bat instead of ./setup.sh - see docs/WINDOWS.md.
• AMD: docs/AMD_MAYA.md.
• Updating: the page's Settings › About and updates, or git pull then ./setup.sh. The model isn't downloaded again.
• Every option (GPUs, context, a model folder you already have, remote access, ...): docs/SETUP.md.

The start takes a few minutes: the engine pins most of the free RAM (all but about 6 GB) for its expert tier, so other programs get little RAM while Maya runs.

Something went wrong?

Run ./maya.sh --report (Windows: START-MAYA.bat --report) right after a slow answer or an error, and attach the maya-report.txt it writes to a GitHub issue. It holds your hardware, your setup and the engine's speed lines; nothing is sent anywhere by itself. The engine's own log is maya-<model>.log in the Maya folder.

• The build stops ("nvcc ... cannot build for these GPUs", "unsupported GNU version", or cuda_profiler_api.h / nvtx3/nvToolsExt.h missing): docs/SETUP.md has the fix for each.
• Slow, with the SSD busy all the time: not enough free RAM for the experts. Close programs or add RAM.
• Windows: answers suddenly several times slower: the card is too full and Windows moves GPU memory to RAM. Leave more of it free: START-MAYA.bat --setup --vram-reserve-mib 2048 (the expert pool's size).
• Windows on a Strix Halo / Gorgon Halo: other programs run out of memory, or the setup warns about the page file: set the page file to the size the setup prints (docs/AMD_MAYA.md).
• The download stopped: run ./maya.sh again; it continues where it stopped.
• Port 8080 is in use: Maya is already running, or --port 8081 starts another one. Settings > Port moves Maya to another port for good (it starts again there).

Support

Project Maya is free and open source. If it's useful to you, you can support it here - thank you!

Speed reports, bug reports and pull requests are welcome: CONTRIBUTING.md says what helps most.

Credits and license

• People: Project Maya is created and led by Peasant Smith (@mw00, X), who owns the project, its name and the Maya models. @needmorevram (X) is a co-maintainer and contributed major features: DFlash2 drafting, EXL3 support, KV streaming, conversation checkpoints, the layer-split search, the expert profile and the CPU expert kernels, and for maya-ds4 the 3.5-bit EXL3 option, tensor parallel over 4 / 8 GPUs and the threaded expert-cache swaps (#276). Thanks also to everyone who sent code and speed reports (contributors).
• Built on Strata (MIT License, Copyright (c) 2026 Niko1221 and the Strata contributors): Maya's engine started from Strata's and was rewritten for GLM-5.3-Flash (the expert tiers across VRAM, RAM and SSD, the layer split, MTP decoding), and its server and dashboard grew from Strata's. And on ggml / llama.cpp (MIT License, Copyright (c) 2023-2026 The ggml authors): the quantization formats, the CPU dot products and the prompt path's MMQ kernels, built from a pinned commit (third_party/ggml/LICENSE).
• EXL3 support follows exllamav3's format, codebooks, CPU expert kernel and tensor-core decoding of a tile (MIT License, Copyright (c) 2025 Turboderp; third_party/exllamav3/LICENSE). The DFlash2 drafter's forward follows beellama.cpp's (MIT License, Copyright (c) 2023-2026 The ggml authors, Anbeeld; third_party/beellama.cpp/LICENSE).
• The models: GLM-5.3-Flash by zai-org (MIT); Maya's quants and the image encoder file are made from zai-org's released weights and keep its license. DeepSeek-V4.1-Flash-1017 by DeepSeek (MIT); Maya's quant, its Engram tables and the drafter are made from DeepSeek's released weights and keep its license; the vision tower comes from DeepSeek's checkpoint, the GGUF option from unsloth's GGUFs, the EXL3 options are LibertAIDAI's (2.9 and 3.5 bpw). The optional DFlash2 drafter (GLM-5.3-Flash-DFlash2-GGUF by Anbeeld) is CC BY-NC-ND 4.0, non-commercial use only, and is downloaded only when you choose it.
• The dashboard's font: Outfit (SIL Open Font License 1.1, serve/web/fonts/OFL.txt); its icons: Lucide (ISC License, serve/web/LICENSE.lucide.txt); its UI library: Preact (MIT License) with htm (Apache License 2.0), bundled in serve/web/preact.js with their notices.
• Maya is open source under the MIT License; the notices of Strata, ggml, exllamav3 and beellama.cpp stay with every copy.