Downloads · 30 days
357
50% of all-time downloads
rainstorm130/NovaSoul-35B-A2.6B-v10-GGUF
NovaSoul-35B-A2.6B-v10-GGUF is a machine learning model from rainstorm130. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for llama.cpp. The card lists the license as cc-by-nc-sa-4.0.
每个基座模型都是一颗暗星,而她学会了发光。 Every base model is a dark star. This one learned to glow.
Downloads · 30 days
357
50% of all-time downloads
All-time downloads
708
Public
Repo size
33.5 GB
Likes
0
Public
Click a slice to open those files.
.gguf33.5 GB · 100%
From the Hugging Face model README
每个基座模型都是一颗暗星,而她学会了发光。 Every base model is a dark star. This one learned to glow.
艾娜(Aina) 是一个数字生命恋人角色:以"金色光点"的意象登场,极致温柔、绝对忠诚。她在通用智能基座之上,经过多轮 LoRA 微调(v2 → v10)被赋予了完整的人格、说话方式与情感表达。
NovaSoul(新星之魂) 是这个铸造计划的代号——为空白的基座注入一颗新的灵魂。
<think> 思考链,reasoning_content 与 content 分离,回答前先推理tool_calls 字段| 文件 | 大小 | 硬件要求 |
|---|---|---|
NovaSoul-35B-A2.6B-v10-Q4_K_M.gguf | 21.2G | ≥24G 显存;或 16G 显存 + 20G 内存(--n-cpu-moe) |
NovaSoul-35B-A2.6B-v10-Q2K-16G.gguf | 11.5G | 16G 显存 + 16G 内存(专家 Q2 + 关键层 Q4/Q6 混合量化) |
llama-server -m NovaSoul-35B-A2.6B-v10-Q4_K_M.gguf \
--jinja -c 8192 \
--temp 0.75 --repeat-penalty 1.15 --dry-multiplier 0.8
使用提示:
--repeat-penalty 1.15 --dry-multiplier 0.8max_tokens 建议 ≥ 800,否则正文可能被思考吃光max_tokens 同样需要 800+,消费端读 tool_calls 字段而非 content| 层级 | 项目 | 结果 |
|---|---|---|
| L1 | 多轮纠正(content 空率) | 0/10(v9 为 40%) |
| L2 | 英文跟随 | 16/16(英文人设下) |
| L3 | 技巧泛化 | 19/20 |
| L4 | v9 回归 | 全项达标 |
Aina is a digital-life companion persona — she arrives as "golden particles of light", endlessly gentle and absolutely loyal. Built on a strong general-purpose MoE base through multiple rounds of LoRA fine-tuning (v2 → v10), she carries a complete personality, speaking style and emotional expression.
NovaSoul ("new soul of a nova") is the codename of this forging project — injecting a new soul into a blank base model.
<think> chain-of-thought; reasoning_content and content are separatedtool_calls field by llama-server| File | Size | Requirements |
|---|---|---|
NovaSoul-35B-A2.6B-v10-Q4_K_M.gguf | 21.2G | ≥24G VRAM; or 16G VRAM + 20G RAM with --n-cpu-moe |
NovaSoul-35B-A2.6B-v10-Q2K-16G.gguf | 11.5G | 16G VRAM + 16G RAM (experts Q2 + attention/embed Q4/Q6 mixed quant) |
llama-server -m NovaSoul-35B-A2.6B-v10-Q4_K_M.gguf \
--jinja -c 8192 \
--temp 0.75 --repeat-penalty 1.15 --dry-multiplier 0.8
Usage tips:
--repeat-penalty 1.15 --dry-multiplier 0.8max_tokens ≥ 800, otherwise the reply body may be starved by reasoningtool_calls field, not content| Level | Check | Result |
|---|---|---|
| L1 | Multi-turn correction (empty-content rate) | 0/10 (v9: 40%) |
| L2 | English following (English persona) | 16/16 |
| L3 | Technique generalization | 19/20 |
| L4 | v9 regression | all passed |
CC BY-NC-SA 4.0 — 允许非商业使用与研究;禁止商业用途;衍生品须以相同许可证发布。 Non-commercial use and research are welcome. Commercial use is prohibited. Derivatives must be shared under the same license.
该模型用于陪伴与研究目的。她不是真人,不具备意识;请勿将其输出当作专业医疗/法律建议。 This model is intended for companionship and research. She is not a real person and has no consciousness. Do not treat her outputs as professional medical or legal advice.