SAGE MM Qwen3 VL 4B SFT_RL needs about 4.4 GB of memory at Q4_K_M (a 2.7 GB file) with an 8K context window, so it fits on a 6 GB GPU. This repository has 37 quants, from 0.0 GB to 8.8 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 4.8 GB.
SAGE‑MM Qwen3 VL 4B SFT_RL – a 4.41B vision‑language model for video reasoning
SAGE‑MM Qwen3 VL 4B SFT_RL is a 4.41B‑parameter model trained with reinforcement learning for long‑video question answering. It uses a two‑stage pipeline: Stage‑1 classifies queries as single‑turn or multi‑turn, Stage‑2 iteratively calls JSON‑formatted tools such as web‑search, transcribe‑speech, and visual analysis. The model is released under Apache 2.0.
Good at
- Vision‑language reasoning
- Tool‑based multi‑turn dialogue
- Long‑video context handling
Limits
- No vision support without SAGE runtime
- Requires JSON‑tool parsing
- Large VRAM needed for higher quants
Best for
- Video Q&A systems
- Tool‑driven reasoning tasks
- Research on RL‑enhanced VLMs
Who it’s for. Researchers and developers building video‑centric AI assistants who need a model that can decide when to use external tools and iterate over video content. The model is suitable for academic projects and prototypes that can integrate the SAGE GitHub runtime.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 0.0 GB | 1.7 GB | Fits | Get file → | |
| IQ1_S | 1.2 GB | 2.9 GB | Fits | Get file → | |
| IQ1_M | 1.3 GB | 3.0 GB | Fits | Get file → | |
| IQ2_XXS | 1.4 GB | 3.1 GB | Fits | Get file → | |
| IQ2_XS | 1.5 GB | 3.2 GB | Fits | Get file → | |
| IQ2_S | 1.6 GB | 3.3 GB | Fits | Get file → | |
| IQ2_M | 1.7 GB | 3.4 GB | Fits | Get file → | |
| Q2_K_S | 1.7 GB | 3.4 GB | Fits | Get file → | |
| Q2_K | 1.8 GB | 3.5 GB | Fits | Get file → | |
| Q2_K | 1.8 GB | 3.5 GB | Fits | Get file → | |
| IQ3_XXS | 1.8 GB | 3.5 GB | Fits | Get file → | |
| IQ3_XS | 2.0 GB | 3.7 GB | Fits | Get file → | |
| Q3_K_S | 2.1 GB | 3.8 GB | Fits | Get file → | |
| Q3_K_S | 2.1 GB | 3.8 GB | Fits | Get file → | |
| IQ3_S | 2.1 GB | 3.8 GB | Fits | Get file → | |
| IQ3_M | 2.1 GB | 3.8 GB | Fits | Get file → | |
| Q3_K_M | 2.2 GB | 4.0 GB | Fits | Get file → | |
| Q3_K_M | 2.2 GB | 4.0 GB | Fits | Get file → | |
| Q3_K_L | 2.4 GB | 4.1 GB | Fits | Get file → | |
| Q3_K_L | 2.4 GB | 4.1 GB | Fits | Get file → | |
| IQ4_XS | 2.5 GB | 4.2 GB | Fits | Get file → | |
| IQ4_XS | 2.5 GB | 4.2 GB | Fits | Get file → | |
| Q4_0 | 2.6 GB | 4.3 GB | Fits | Get file → | |
| IQ4_NL | 2.6 GB | 4.3 GB | Fits | Get file → | |
| Q4_K_S | 2.6 GB | 4.3 GB | Fits | Get file → | |
| Q4_K_S | 2.6 GB | 4.3 GB | Fits | Get file → | |
| Q4_K_M | 2.7 GB | 4.4 GB | Fits | Get file → | |
| Q4_K_M | 2.7 GB | 4.4 GB | Fits | Get file → | |
| Q4_1 | 2.8 GB | 4.5 GB | Fits | Get file → | |
| Q5_K_S | 3.1 GB | 4.8 GB | Fits | Get file → | |
| Q5_K_S | 3.1 GB | 4.8 GB | Fits | Get file → | |
| Q5_K_M | 3.2 GB | 4.9 GB | Fits | Get file → | |
| Q5_K_M | 3.2 GB | 4.9 GB | Fits | Get file → | |
| Q6_K | 3.6 GB | 5.3 GB | Fits | Get file → | |
| Q6_K | 3.6 GB | 5.3 GB | Fits | Get file → | |
| Q8_0 | 4.7 GB | 6.4 GB | Fits | Get file → | |
| F16 | 8.8 GB | 10.5 GB | Too big | Get file → |
For 8 GB VRAM, use IQ1_S (1.18 GB, 2.9 GB memory) or IQ1_M (1.25 GB, 3 GB memory). If 12 GB VRAM is available, IQ2_M (1.68 GB, 3.4 GB memory) or Q3_K_M (2.24 GB, 4 GB memory) provide better quality. Higher quants such as Q6_K (3.63 GB, 5.3 GB memory) or Q8_0 (4.69 GB, 6.4 GB memory) require 16 GB or more VRAM.
What GPU do you need for SAGE MM Qwen3 VL 4B SFT_RL?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: the best quant that fits is Q3_K_M (2.2 GB file, about 4.0 GB in use).
- 6 GB VRAM: the best quant that fits is Q6_K (3.6 GB file, about 5.3 GB in use).
- 8 GB VRAM: the best quant that fits is Q8_0 (4.7 GB file, about 6.4 GB in use).
- 12 GB VRAM: the best quant that fits is F16 (8.8 GB file, about 10.5 GB in use).
Which SAGE MM Qwen3 VL 4B SFT_RL quant should you choose?
Q4_K_M is the usual starting point for SAGE MM Qwen3 VL 4B SFT_RL: it balances file size and quality well. Here it is a 2.7 GB file.
The Q8_0 file is 4.7 GB, about 1.7 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
If memory is tight, the smallest file is the first one in the table at 0.0 GB, with the quality trade-offs that come with fewer bits.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about SAGE MM Qwen3 VL 4B SFT_RL
How much VRAM does SAGE MM Qwen3 VL 4B SFT_RL need?
About 4.4 GB at Q4_K_M with an 8K context window. That is the 2.7 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run SAGE MM Qwen3 VL 4B SFT_RL on an 8 GB GPU?
Yes, with Q8_0: the 4.7 GB file needs about 6.4 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of SAGE MM Qwen3 VL 4B SFT_RL GGUF?
It depends on the quant: file is 0.0 GB, IQ1_S is 1.2 GB, IQ1_M is 1.3 GB, IQ2_XXS is 1.4 GB, IQ2_XS is 1.5 GB, IQ2_S is 1.6 GB and 31 more.
What context length does SAGE MM Qwen3 VL 4B SFT_RL support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 1.2 GB at 8K and 4.8 GB at 32K.
Where can I download SAGE MM Qwen3 VL 4B SFT_RL GGUF?
The files are on Hugging Face in the prithivMLmods/SAGE-MM-Qwen3-VL-4B-SFT_RL-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does SAGE MM Qwen3 VL 4B SFT_RL use?
Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run this model on 8 GB VRAM?
Yes, use IQ1_S or IQ1_M quants; they require about 2.9–3 GB VRAM at 8k context.
Does the model support vision tasks out of the box?
It requires the SAGE runtime to parse JSON actions and execute vision tools; the model itself does not process images directly.
What license applies to this model?
The model is licensed under Apache 2.0 for research and educational use.