spark 1.5 . by forgewright-ai . github.com/forgewright-ai/spark
~ > spark model list -- 26 models, one list
| model | file | RAM | license | tested | note |
|---|---|---|---|---|---|
| qwen3-1-7b | 1.0 GB | 3 GB | Apache 2.0 | line | |
| qwen3-4b | 2.3 GB | 5 GB | Apache 2.0 | line | |
| qwen3-8b | 4.7 GB | 7 GB | Apache 2.0 | line | |
| qwen3-14b | 8.4 GB | 11 GB | Apache 2.0 | line | |
| qwen3-30b-a3b | 17.4 GB | 21 GB | Apache 2.0 | line | mixture of experts: near-4B speed, near-flagship answers -- 32 GB machines |
| qwen3-4b-thinking | 2.3 GB | 5 GB | Apache 2.0 | thinks before it answers: slower, for hard questions -- a second model | |
| qwen3-coder-30b-a3b | 17.3 GB | 21 GB | Apache 2.0 | code, mixture of experts -- 32 GB machines | |
| qwen2-5-coder-7b | 4.4 GB | 7 GB | Apache 2.0 | code: reads and writes programs, explains errors | |
| qwen2-5-7b | 4.4 GB | 7 GB | Apache 2.0 | ||
| qwen2-5-14b | 8.4 GB | 11 GB | Apache 2.0 | ||
| mistral-7b-v0-3 | 4.1 GB | 6 GB | Apache 2.0 | ||
| mistral-nemo-12b | 7.0 GB | 10 GB | Apache 2.0 | ||
| phi-4-mini | 2.3 GB | 5 GB | MIT | ||
| phi-4 | 8.4 GB | 11 GB | MIT | ||
| deepseek-r1-qwen-7b | 4.4 GB | 7 GB | MIT | thinks before it answers: slower, for hard questions | |
| deepseek-r1-qwen-14b | 8.4 GB | 11 GB | MIT | thinks before it answers: slower, for hard questions | |
| smollm2-1-7b | 1.0 GB | 3 GB | Apache 2.0 | ||
| gpt-oss-20b | 10.8 GB | 14 GB | Apache 2.0 | mixture of experts, thinks before it answers -- 16 GB machines and up | |
| granite-3-3-8b | 4.6 GB | 7 GB | Apache 2.0 | ||
| llama3-2-1b | 0.8 GB | 3 GB | Llama 3.2 Community License | not an open-source license -- read the terms before you download | |
| llama3-2-3b | 1.9 GB | 4 GB | Llama 3.2 Community License | not an open-source license -- read the terms before you download | |
| llama3-1-8b | 4.6 GB | 7 GB | Llama 3.1 Community License | not an open-source license -- read the terms before you download | |
| gemma3-1b | 0.8 GB | 3 GB | Gemma Terms of Use | not an open-source license -- read the terms before you download | |
| gemma3-4b | 2.3 GB | 5 GB | Gemma Terms of Use | not an open-source license -- read the terms before you download | |
| gemma3-12b | 6.8 GB | 9 GB | Gemma Terms of Use | not an open-source license -- read the terms before you download | |
| gemma3-27b | 15.4 GB | 19 GB | Gemma Terms of Use | not an open-source license -- read the terms before you download |
spark model NAME serves one; spark ember NAME adds a second, for conversations. auto picks among the rows tested on the line under Apache-2.0 or MIT; any other license is shown and asked about before the download. Every file is sha256-verified. Your own rows: spark model add URL --license. A row is a pull request: CONTRIBUTING.md.