Chandra — Hardware Requirements & GPU Compatibility
Vision
Chandra is Datalab's 8.8-billion-parameter vision-language OCR model, built on a Qwen3VL backbone, that converts document images and PDFs into markdown, HTML, or JSON while preserving layout, tables, math, and forms with checkboxes. It handles handwriting, multi-column and complex layouts, and extracts images and diagrams with captions and structured data across more than 40 languages, aiming at document digitization rather than general chat. On the olmOCR benchmark Chandra scored highest overall among compared models, ahead of Datalab's own Marker pipeline, Mistral's OCR API, DeepSeek-OCR, and anchored GPT-4o and Gemini Flash 2 baselines. At under 9 billion parameters it runs on a single consumer GPU. Context length is 262,144 tokens. It is released under an OpenRAIL license, a responsible-AI license that permits broad use but restricts certain harmful applications. It was published in October 2025 and has since been superseded by a newer Chandra OCR 2 model.
87.1K downloads
532 likesSpecifications
- Publisher
- datalab-to
- Parameters
- 8.8B
- Architecture
- Qwen3VLForConditionalGeneration
- Context Length
- 262,144 tokens
- Vocabulary Size
- 151,936
- Release Date
- 2025-10-21
- License
- openrail
Get Started
HuggingFace
How Much VRAM Does Chandra Need?
Select a quantization to see compatible GPUs below.
| Quantization | Bits | VRAM | + Context | File Size | Quality |
|---|---|---|---|---|---|
| Q2_Kest. | 3.40 | 4.3 GB | 42.7 GB | 3.73 GB | 2-bit quantization with K-quant improvements |
| Q3_K_Mest. | 3.90 | 4.9 GB | 43.2 GB | 4.27 GB | 3-bit medium quantization |
| Q4_K_Mest. | 4.80 | 5.9 GB | 44.2 GB | 5.26 GB | 4-bit medium quantization — most popular sweet spot |
| Q5_K_Mest. | 5.70 | 6.8 GB | 45.2 GB | 6.25 GB | 5-bit medium quantization — good quality/size tradeoff |
| Q6_Kest. | 6.60 | 7.8 GB | 46.2 GB | 7.23 GB | 6-bit quantization, very good quality |
| Q8_0est. | 8.00 | 9.4 GB | 47.7 GB | 8.77 GB | 8-bit quantization, near-lossless |
| BF16est. | 16.00 | 18.1 GB | 56.5 GB | 17.53 GB | Brain floating point 16 — preferred for training |
est.= calculated VRAM estimate; no published GGUF file found for that quantization yet. Other rows are verified against real community uploads.
Which GPUs Can Run Chandra?
Q4_K_M · 5.9 GBChandra (Q4_K_M) requires 5.9 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 8+ GB is recommended. Using the full 262K context window can add up to 38.4 GB, bringing total usage to 44.2 GB. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 3070 Ti.
Runs great
— Plenty of headroom
Decent
— Enough VRAM, may be tightWhich Devices Can Run Chandra?
Q4_K_M · 5.9 GB58 devices with unified memory can run Chandra, including NVIDIA DGX H100, NVIDIA DGX A100 640GB, MacBook Air 13" M3 (8 GB).
Runs great
— Plenty of headroomRelated Models
Frequently Asked Questions
- How much VRAM does Chandra need?
Chandra requires 5.9 GB of VRAM at Q4_K_M, or 18.1 GB at BF16. Full 262K context adds up to 38.4 GB (44.2 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 8.8B × 4.8 bits ÷ 8 = 5.3 GB
KV Cache + Overhead ≈ 0.6 GB (at 2K context + ~0.3 GB framework)
KV Cache + Overhead ≈ 38.9 GB (at full 262K context)
VRAM usage by quantization
Q4_K_M5.9 GBQ4_K_M + full context44.2 GB- What's the best quantization for Chandra?
For Chandra, Q4_K_M (5.9 GB) offers the best balance of quality and VRAM usage. Q5_K_M (6.8 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 4.3 GB.
VRAM requirement by quantization
Q2_K4.3 GBQ4_K_M ★5.9 GBQ5_K_M6.8 GBQ6_K7.8 GBQ8_09.4 GBBF1618.1 GB★ Recommended — best balance of quality and VRAM usage.
- Can I run Chandra on a Mac?
Chandra requires at least 4.3 GB at Q2_K, which exceeds the unified memory of most consumer Macs. You would need a Mac Studio or Mac Pro with a high-memory configuration.
- Can I run Chandra locally?
Yes — Chandra can run locally on consumer hardware. At Q4_K_M quantization it needs 5.9 GB of VRAM. Popular tools include Ollama, LM Studio, and llama.cpp.
- How fast is Chandra?
At Q4_K_M, Chandra can reach ~819 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~112 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: NVIDIA B200 → 8000 ÷ 5.9 × 0.65 = ~887 tok/s
Estimated speed at Q4_K_M (5.9 GB)
~887 tok/s~112 tok/s~887 tok/s~819 tok/sReal-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
- What's the download size of Chandra?
At Q4_K_M, the download is about 5.26 GB. The full-precision BF16 version is 17.53 GB. The smallest option (Q2_K) is 3.73 GB.
- Which GPUs can run Chandra?
52 consumer GPUs can run Chandra at Q4_K_M (5.9 GB). Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT, AMD Radeon RX 7600. 40 GPUs have plenty of headroom for comfortable inference.
- Which devices can run Chandra?
59 devices with unified memory can run Chandra at Q4_K_M (5.9 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Apple iPhone 17 Pro, Asus ROG Flow Z13 (2025, Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.