Setting up a local coding agent - Revision history

Ajshen: /* 18. Reference deployment */

2026-09-04T04:52:12Z

18. Reference deployment

← Older revision		Revision as of 04:52, 4 September 2026
Line 973:		Line 973:

	== 18. Reference deployment ==		== 18. Reference deployment ==
			Alex Shen (ajshen) has a successful working deployment. Feel free to contact him with any questions.
	{\| class="wikitable"		{\| class="wikitable"
	\|-		\|-

Ajshen at 04:48, 4 September 2026

2026-09-04T04:48:46Z

Show changes

Ajshen at 04:37, 4 September 2026

2026-09-04T04:37:53Z

← Older revision		Revision as of 04:37, 4 September 2026
Line 8:		Line 8:
	hardware.		hardware.

	It is written to work on any single-GPU server. Everything specific to the '''Wright'''		It is written to work on any single-GPU server. Everything specific to the '''[[Wright GPU Cluster\|Wright]]'''
	cluster at CMU is called out in boxes like this one:		cluster at CMU is called out in boxes like this one:

Line 19:		Line 19:
	</blockquote>		</blockquote>

	Skip to [[#13. Restarting after a new allocation\|Restarting after a new allocation]] if the		Skip to [[Setting up a local coding agent#13. Restarting after a new allocation\|Restarting after a new allocation]] if the
	software is already installed and you just need the server back up.		software is already installed and you just need the server back up.

Line 149:		Line 149:
	'''Version floor: vLLM 0.28.0.''' The model card states a lower floor (0.17.0), but 0.28.0		'''Version floor: vLLM 0.28.0.''' The model card states a lower floor (0.17.0), but 0.28.0
	is what this configuration is proven against, and the optional speculative-decoding path in		is what this configuration is proven against, and the optional speculative-decoding path in
	[[#12. Optional: MTP speculative decoding\|§12]] requires it. Do not go below it.		[[Setting up a local coding agent#12. Optional: MTP speculative decoding\|§12]] requires it. Do not go below it.

	Verify when both finish:		Verify when both finish:

Ajshen: Created page with "= Deploying Qwen3.8-27B with OpenCode on a single 48 GB GPU = This page is a complete, standalone recipe for standing up an OpenAI-compatible [https://docs.vllm.ai vLLM] endpoint serving `Qwen/Qwen3.8-27B-FP8` on '''one''' 48 GB Ada-generation GPU (L40, L40S, or RTX 6000 Ada), and wiring [https://opencode.ai OpenCode] to it as an agentic coding CLI. When you are done you have a local, private coding assistant with a 128k context window that runs entirely on y..."

2026-09-04T04:37:30Z

Created page with "= Deploying Qwen3.8-27B with OpenCode on a single 48 GB GPU = This page is a complete, standalone recipe for standing up an OpenAI-compatible [https://docs.vllm.ai vLLM] endpoint serving <code>Qwen/Qwen3.8-27B-FP8</code> on '''one''' 48 GB Ada-generation GPU (L40, L40S, or RTX 6000 Ada), and wiring [https://opencode.ai OpenCode] to it as an agentic coding CLI. When you are done you have a local, private coding assistant with a 128k context window that runs entirely on y..."

Show changes