schema = "m64/capsule@1" description = "opencode on Qwen3.8-Flash-Next IQ3_S, local (Strata, GB10). opencode 1.18.23 (sst/opencode), driven through `opencode serve`, with Qwen3.8-Flash-Next (Qwen's 125B-A6B MoE, ISTA-DASLab's GSQ-RCO IQ3_S GGUF) served by Strata 0.2.13's DGX Spark engine in the same container: 262144-token context, MTP drafting, pictures on. Each session gets an empty directory; the agent can write and run code there, and every tool call is reported. Asks for no network: prompts and code stay on this machine." kinds = ["agent.session"] authors = ["dh7"] license = "mit" [hardware] platform = "linux-aarch64" gpu = "NVIDIA GB10" gpu_count = 1 # Unified memory: Strata reads the experts in place from shard 1 (55 GB) and locks the 29 GB n-gram table in # RAM, plus the pack, the draft layer, the vision tower and a 262144-token int8 KV cache (12 of the 48 layers # keep one); the harness and gateway need about 2 GB more. vram_gb = 82 driver = ">=580.65" disk_gb = 92 [runtime] sandbox = "container" # No image: the Dockerfile builds the qwen38-flash-next-strata-gb10 capsule's engine image plus Node 22 and opencode. gpu = true shm_gb = 2 internal_port = 9500 health = "/health" expected_load_seconds = 900 user = "65534:65534" # opencode's own server (web UI + full API, SSE and the terminal's WebSocket) on a second # front port. Raw: no per-session isolation and no params filtering there. [runtime.app] internal_port = 4096 # ISTA-DASLab's GSQ-RCO shards at the revision Strata 0.2.13 pins (setup.py HF_REVISIONS), and the picture reader. [[artifacts]] type = "hf" repo = "ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-GGUF" revision = "ed59f92082b1e93c0e96d60a8b11aab089b52f09" include = ["README.md", "IQ3_S/*.gguf", "mmproj-Qwen3.8-Flash-Next-BF16.gguf"] mount = "/weights/qwen38fn"