Naive-N0.5 Guide: VRAM Requirements
About This Model
Naive-N0.5-Flash is a Mixture-of-Experts (MoE) model built for coding and AI research and development, featuring 309B total parameters and 15.5B active parameters. It adopts the open-weight MiMo-V2.5 as its base model.
This model does not have full attention layers. Instead, it employs a hybrid attention mechanism combining Sliding Window Attention (SWA) and lightweight DeepSeek Sparse Attention (DSA), supporting a native 1M token context window. Out of all 48 layers, 39 are SWA layers (with a window width of 128 tokens) and 9 are DSA layers (selecting the top 2,048 tokens, GQA4).
What Makes It Stand Out
- By combining SWA and DSA without using full attention layers, it can process a native 1M token context while reducing computational cost and memory consumption.
- In the dedicated inference system NaiveRT, it delivers speeds of 50 tokens/s per user in Standard mode and up to 2,000 tokens/s in Ultrafast mode.
- It has undergone multi-stage training spanning 3.25T tokens (Indexer Warmup, Sparse Attention Training, and Learning Rate Decay) under a native 1M token context window environment.
Caveats and Weaknesses
- Although backbone computations are reduced by sparse attention, the indexer must scan the entire history, requiring the maintenance of a complete KV cache.
- The published evaluation scores were measured under specific conditions, such as the Claude Code 2.1.207 environment (temperature 1.0, top-p 0.95, basic file I/O and Bash tools only) and an in-house AutoResearch harness.
Running It Locally
- Since the model weights alone occupy approximately 315 GB, an NVIDIA GPU environment with FP8 support and additional GPU memory for inference is required.
- For general use, sampling parameters of temperature=1.0 and top_p=0.95 are recommended.
- FP8 mixed-precision inference code using transformers is provided, with transformers version 5.17.0 or higher specified.
- License
mit(Commercial use allowed): Permits commercial use, modification and redistribution, provided the copyright notice and license text are retained.
This license summary is not legal advice; check the original terms before use.
Sources: model cards for NaiveAI/Naive-N0.5-Flash, as of 2026-10-05. This explanation was drafted by AI from those cards and checked by code and a second model before publication.
Our Coverage and Data
Everything Local Model Watch has published about the Naive-N0.5 family: 1 article(s) covering the base model and its fine-tunes, plus converted builds we tracked after publication. Memory requirements below are computed by this site from file sizes, not quoted from model cards. Part of our model family index.
At a Glance
| Item | Value |
|---|---|
| Base model(s) | NaiveAI/Naive-N0.5-Flash |
| Publisher | NaiveAI |
| License (model card) | mit |
| Articles | 1 |
Hardware Requirements
Estimated requirements (calculated by Local Model Watch)
| Your VRAM | Quantization | File size | Est. memory needed |
|---|---|---|---|
| More than 690GB of VRAM (multi-GPU or CPU offload required) | Original precision | 575.4GB | 690.5GB |
Memory estimates add a 20% runtime overhead (KV cache, etc.) to the actual size of the distributed files. Actual usage varies with context length, batch size and inference engine. These figures are computed by this site from file sizes, not published by the model’s authors. Compare with other models in our VRAM quick reference. What the quantization names mean: glossary.
Can You Run It Locally?
Not usable in Ollama, LM Studio and llama.cpp yet.
The publisher ships safetensors only, and llama.cpp’s registry does not list this architecture. llama.cpp would need to add support before these tools can run it. 4 converted build(s) from other uploaders exist. Today it can be run with transformers, using the memory figures in the table above.
License — mit (Commercial use allowed): Permits commercial use, modification and redistribution, provided the copyright notice and license text are retained.
Compiled by this site’s code from the published formats, converted builds we have found, and each engine’s own model registry. “Not found" means we have not seen such a build, not that none exists. License summaries are not legal advice — check the publisher’s original terms before relying on them.
Quantized and Converted Variants
→ Scroll horizontally to see all columns
| Added | Publisher | Format | Repository | Smallest VRAM tier (build, est. memory) |
|---|---|---|---|---|
| 2026-10-05 | NaiveAI | FP8 | NaiveAI/Naive-N0.5-Flash-FP8-Draft | FP8 1.5GB (fits in 4GB VRAM) |
| 2026-10-05 | NaiveAI | FP8 | NaiveAI/Naive-N0.5-Flash-FP8 | FP8 352.2GB (does not fit a single consumer GPU) |
File sizes of each build:
- Available builds in NaiveAI/Naive-N0.5-Flash-FP8-Draft: FP8 1.2GB
- Available builds in NaiveAI/Naive-N0.5-Flash-FP8: FP8 293.5GB
This section is appended automatically by Local Model Watch when a converted build of this model appears after publication. Memory figures are estimated from the size of the distributed files.
Articles (the family’s own models first, then newest)
| Published | Model | Type | Article |
|---|---|---|---|
| 2026-10-05 | NaiveAI/Naive-N0.5-Flash | New Models | Naive-N0.5-Flash MoE Model for AI Research and Coding: ~690GB Memory |
Repositories
Last updated 2026-11-22 (JST). The explanation at the top of this page was written with the help of AI from the primary sources it cites. The tables and lists under “Our Coverage and Data" are assembled by code from our article log.