Ornith-1.5-9B vs Qwen3-Coder-30B-A3B-Instruct

Ornith-1.5-9B

A small coding and terminal-agent cousin with thinking on by default, XML tool calls that vLLM and SGLang parse, and 262,144 tokens of context. Ornith's own 4-bit builds run in under 8 GB in most local runtimes. The main risk is an absent publisher, with no licence file, no safety results, files replaced without a note and no replies.

Qwen3-Coder-30B-A3B-Instruct

A fast coding cousin, with 3.3B of 30.5B parameters active per token, for agentic coding tools, repository reading and fill-in-the-middle without a thinking step. It needs about 21 to 23 GB at 4 bits, and several providers host it. The main risk is that it runs commands for agents with no safety results or guidance, and Alibaba doesn't answer users.

Ornith-1.5-9B

1. Start it with ollama run ornith-1.5:9b, or llama-server -hf ornith-ai/Ornith-1.5-9B-GGUF with the Q4_K_M file
2. In vLLM pass --enable-auto-tool-choice --tool-call-parser qwen3_xml --reasoning-parser qwen3, and in SGLang --tool-call-parser qwen3_coder --reasoning-parser qwen3, so tool calls and reasoning come back parsed
3. Thinking is on by default. Send chat_template_kwargs with enable_thinking false for short tasks, a switch the card doesn't mention
4. Pin which template your runtime loads, since chat_template.jinja keeps earlier reasoning and the tokenizer_config.json template drops it
5. Use temperature 0.6, top_p 0.95, top_k 20 for coding and temperature 1.0 with presence_penalty 1.5 for general tasks, per the card
6. Pin a revision when you download, since Ornith has replaced files in place, and add mmproj-Ornith-1.5-9B-BF16.gguf (0.9 GB) for image input in llama.cpp
7. Validate JSON output yourself, since structured output isn't documented

Qwen3-Coder-30B-A3B-Instruct

1. Quickest start is ollama run qwen3-coder:30b, 19 GB
2. On our reference machine use BF16 with vllm serve Qwen/Qwen3-Coder-30B-A3B-Instruct --max-model-len 262144 --enable-auto-tool-choice --tool-call-parser qwen3_coder, 61 GB of weights fits the GPU share
3. Use the qwen3_coder parser in vLLM or SGLang, not hermes, since calls come back as function and parameter blocks rather than JSON
4. Pass tool-call arguments in the history as objects, the template iterates them as a mapping
5. Sampling per the card is temperature 0.7, top_p 0.8, top_k 20 and repetition_penalty 1.05, with up to 65,536 output tokens
6. For code completion use the fill-in-the-middle tokens <|fim_prefix|>, <|fim_suffix|> and <|fim_middle|> from Qwen's README
7. Sandbox the commands and files it touches, since nothing from Alibaba covers prompt injection