docs: plan Flash-Next tiered inference for RTX 4070 Ti SUPER and 32GB RAM

2026-09-21 01:40:30 +08:00
commit 0a2c1afb26
8 changed files with 261 additions and 0 deletions
@@ -0,0 +1,23 @@
+{
+  "status": "planning_only_not_hardware_validated",
+  "model_family": "Qwen3.8-Flash-Next",
+  "reference_checkpoint": "nvidia/Qwen3.8-Flash-Next-NVFP4",
+  "gpu": "NVIDIA GeForce RTX 4070 Ti SUPER",
+  "gpu_architecture": "Ada",
+  "vram_gib": 16,
+  "inference_vram_budget_gib": 14,
+  "host_ram_gib_assumed": 32,
+  "process_tree_and_page_cache_budget_gib": 24,
+  "ssd_capacity_tb_decimal": 2,
+  "cpu": null,
+  "ssd_model": null,
+  "memory_channels_and_speed": null,
+  "planned_os": "native Linux x86_64, version to be fixed",
+  "context_tokens": 4096,
+  "max_output_tokens": 512,
+  "concurrency": 1,
+  "mtp_enabled": false,
+  "required_modalities": ["text", "image", "video"],
+  "runtime_revision": null,
+  "deployment_ready": false
+}