AI 模型類型：地圖與名詞表
LLM 或 VLM、開放權重或開源、base、instruct、reasoning、蒸餾、MoE、GGUF 與量化 — 模型名稱中這些字詞的意思，附即時範例。
模型類型
開放程度
架構
檔案格式
量化
| 分組 | 意思 | 範例 | |
|---|---|---|---|
| LLMlarge language model · text model | 模型類型 | A model that reads and writes text: chat, writing, code, analysis. Everything else in this glossary is a flavour or a packaging of it. | |
| VLMVL · vision-language · multimodal | 模型類型 | An LLM that also takes images (screenshots, photos, documents) as input. Locally it usually needs a second file — the vision projector (mmproj) — next to the GGUF weights. | |
| Open weightsopen model · downloadable weights | 開放程度 | The weights can be downloaded and run on your own hardware, under any licence — some (Llama, Gemma) limit commercial use or require accepting terms first. Training data and code usually stay closed. | |
| Open Source AIOSAID · OSI definition | 開放程度 | The stricter OSI definition (OSAID 1.0): weights plus the training code and enough information about the data to rebuild the model, all under open licences. Few models qualify; Apache-2.0 or MIT weights alone are open weights, not necessarily open source. | — |
| Proprietary (API only)closed · API model | 開放程度 | No weights at all: the model runs only on the vendor's servers through an app or a paid API. It cannot be run locally. | — |
| Basepretrained · -Base | 版本變體 | The raw pretrained model: it continues text but does not follow instructions. A starting point for fine-tuning, not for chatting. | — |
| Instruct-it · -Chat · -Instruct | 版本變體 | The base model tuned to follow instructions and hold a dialogue. This is the version you want for a local chat; Google marks it -it, others -Instruct or -Chat. | |
| ReasoningThinking · R1 · -Thinking | 版本變體 | Trained to write out a chain of thought before the answer. Better at maths, code and logic, but spends many more tokens — and so more time — per reply. | |
| Coder-Coder · code model | 版本變體 | Further trained on source code: completion, refactoring, agentic coding in the IDE. General chat quality may be lower than the sibling instruct model. | |
| Distill-Distill · R1-Distill-Qwen | 版本變體 | A smaller model taught on the answers of a bigger one. DeepSeek-R1-Distill-Qwen-32B is a Qwen 32B that imitates R1 — not R1 itself, and far weaker than the 671B original. | — |
| Abliterated / uncensoreduncensored · abliterated · heretic 無安全過濾 | 版本變體 | A community modification with the refusal behaviour removed from the weights. It answers anything, including harmful requests, often with lower quality and no safety guarantees. For personal and research use only; you are responsible for how you use it. | — |
| Densedense model | 架構 | Every parameter works on every token. Speed is set by the full size: a 70B dense model reads all 70B weights from memory for each generated token. | |
| MoEmixture of experts | 架構 | The layers are split into many experts and a router picks a few of them per token. All experts must sit in memory (VRAM + RAM), but each token touches only the active part — so a big MoE runs much faster than a dense model of the same size, and the experts can be offloaded to system RAM. | |
| Active parameters (-A3B)-A3B · -A22B · active | 架構 | In MoE names the suffix gives the parameters used per token: Qwen3-30B-A3B has 30B in total and 3B active. Memory follows the total, speed follows the active number. | |
| Safetensors (BF16)safetensors · BF16 · FP16 | 檔案格式 | The original release format on Hugging Face, usually 16 bits per weight: about 2 GB per billion parameters. Used by vLLM, Transformers and as the source for every quantisation. | — |
| GGUF.gguf · llama.cpp | 檔案格式 | A single-file format of llama.cpp with the weights already quantised. Runs on CPU, GPU or both at once; LM Studio, Ollama and Jan use it. Big models come split into parts (-00001-of-00003). | — |
| AWQ / GPTQ / EXL2-3AWQ · GPTQ · EXL2 · EXL3 | 檔案格式 | Quantised formats for GPU-only servers (vLLM, ExLlama, TGI). Fast when the whole model fits in VRAM; no offload to system RAM. | — |
| MLXmlx-community | 檔案格式 | Apple's framework and weight format for M-series Macs, using the unified memory. Often a little faster on a Mac than GGUF of the same size. | — |
| Q4_K_M and other K-quantsQ4_K_M · Q5_K_M · Q6_K · Q8_0 | 量化 | GGUF quant names: the number is roughly the bits per weight, K marks the k-quant method, S/M/L the mix inside. Q4_K_M (≈4.8 bits) is the usual sweet spot; Q8_0 is almost lossless; below Q3 quality drops noticeably. | — |
| IQ quants and Unsloth Dynamic (UD)IQ2_XXS · IQ3_K · UD-Q2_K_XL | 量化 | Newer low-bit GGUF methods: IQ (importance-matrix) quants and Unsloth Dynamic keep the sensitive layers at higher precision, so 2–3-bit files of huge models stay usable. | |
| FP8 / MXFP4 / NVFP4FP8 · MXFP4 · NVFP4 | 量化 | Low-precision floating-point formats with hardware support in recent GPUs. gpt-oss ships natively in MXFP4 (≈4.25 bits per weight), which is why gpt-oss-120b fits in about 65 GB. |
標示為「無安全過濾」的模型（uncensored / abliterated）僅供個人與研究用途：它們沒有安全限制，使用責任由使用者自行承擔。
更新日期 · 來源: fedi.software (CC BY 4.0), Open Source Initiative (OSAID 1.0)
名詞表分成哪幾組？
模型名稱裡常見的詞分成六組，每個術語都附上實際模型範例，並連到模型比較：
- 模型類型：處理文字的 LLM、也能看圖的 VLM；
- 開放程度：開放權重、Open Source AI、專有；
- 版本變體：base、instruct、reasoning、coder、distill、abliterated / uncensored；
- 架構：稠密（dense）、MoE、活躍參數；
- 檔案格式：Safetensors、GGUF、AWQ / GPTQ / EXL、MLX；
- 量化：Q4_K_M 等 K-quants、IQ 與 Unsloth Dynamic、FP8 / MXFP4。
模型名稱要怎麼讀？
從前往後拆：先是系列與規模，MoE 會接著標活躍參數（例如 -A3B），然後是 -Instruct 或 -it 之類的變體，最後是下載檔案的格式與量化。蒸餾（distill）模型是用大模型的回答訓練出來的小模型，並不是大模型本身。
本機聊天該挑哪種模型？
先選 Instruct（或 reasoning）變體，再確認手上的記憶體，挑放得下的量化，接著決定需要的上下文長度，最後到哪個開放模型適合您的硬體確認能接受的速度。Q4_K_M 是大小與品質之間常見的平衡點，Q8_0 幾乎無損，低於 Q3 品質會明顯下降。
開放權重就是開源嗎？
不是。開放權重（open weights）可能附帶限制用途的授權，訓練資料通常也不公開；判斷標準可參考 OSI 的 Open Source AI Definition 1.0。來源還包括 Hugging Face 的模型卡與設定檔，以及 llama.cpp 文件。
abliterated / uncensored 是什麼？
這類模型移除了拒答行為。名詞表只以中立方式列出，不作推薦；僅限個人與研究用途，使用責任由使用者自行承擔。