Add server settings for limited memory

Document existing public MLX-LM cache and concurrency options and their synthetic multi-turn checks. No third-party log details are included. Model files and runtime code remain unchanged; full-model capacity is not established by these tests.

TROUBLESHOOTING.md

@@ -43,6 +43,36 @@ separate runtime. GUI integration has not been validated for these full checkpoi
 Do not remove vision/MTP parameters or change the config to force a stock text
 loader to accept the file. All 2,379 saved tensor entries belong to this release.
 
+## Server settings for limited memory
+
+The pinned MLX-LM server retains prompt caches for reuse by default. For a setup
+that favors lower memory use over reuse speed, stop the existing server and start
+it from the same patched Python environment with:
+
+```bash
+mlx_lm.server --model ukisai/Swift-1.5-4bit-MLX \
+  --host 127.0.0.1 --port 8080 \
+  --prompt-cache-size 0 \
+  --prompt-concurrency 1 --decode-concurrency 1 \
+  --prefill-step-size 512
+```
+
+This disables retained prompt caches, limits concurrency to one, and processes
+prefill in smaller chunks. The model weights and their quantization are unchanged.
+The active request still needs its own cache and temporary memory; these options
+do not set a total process-memory cap or guarantee that any context length fits.
+
+Disabling reuse can slow later turns because their history must be processed
+again. It does not remove history that the client includes in the next request.
+Start with a fresh short conversation and increase history while monitoring RAM.
+The model's configured context limit is distinct from the memory needed to run it.
+
+Validation scope: three successive HTTP requests, including a streaming request,
+passed on each small synthetic 4-bit and 5-bit architecture with these settings.
+Retained-cache accounting stayed at zero. These tests did not load the complete
+Swift checkpoint or establish full-model long-context capacity on any Mac.
+No macOS or MLX memory-limit overrides were used in those tests.
+
 ## 3. If a server reports `404 generation thread died`
 
 In the pinned official MLX-LM server, this message means its generation worker
@@ -75,5 +105,5 @@ against a dead worker will not recover it; restart only after addressing the cau
   a tool definition. These are architecture/server checks, not full Swift generation
   or a test of the reported third-party plugin.
 
-The reporter's original traceback is still needed to identify their specific failure.
 These checks do not establish full-model Mac execution or GUI compatibility.
+Diagnosing an individual crash requires its original exception and runtime configuration.

UPLOAD_MANIFEST.json

@@ -36,13 +36,13 @@
     },
     {
       "path": "TROUBLESHOOTING.md",
-      "bytes": 4360,
-      "sha256": "6460abf4aa4d2de46f07ce8dfa095476467eb466c6250c727ba0d6f2c77399eb"
+      "bytes": 5845,
+      "sha256": "461ca48d8ec617a226f5c1d726c41bf5dfba2121fe8e7c4ad389d39fb1ccec4c"
     },
     {
       "path": "USAGE.md",
-      "bytes": 5924,
-      "sha256": "70e4b505d7acb75d33774e19b4b6202e5e45f0dcaf0f5973d6ea73f5d618a941"
+      "bytes": 7405,
+      "sha256": "51f0a3c01149644dfc17db8b08c4b02c4de485e871a13c53e361165718711618"
     },
     {
       "path": "chat_template.jinja",
@@ -280,8 +280,9 @@
       "sha256": "ce99b4cb2983d118806ce0a8b777a35b093e2000a503ebde25853284c9dfa003"
     }
   ],
-  "total_bytes_excluding_manifest": 15859840103,
+  "total_bytes_excluding_manifest": 15859843069,
   "manifest_self_excluded": true,
   "packaging_update": "2026-09-24: diagnostic fixtures archived; model weights unchanged",
-  "diagnostic_update": "2026-09-25: clarify required runtime; add download/runtime checker; weights unchanged"
+  "diagnostic_update": "2026-09-25: clarify required runtime; add download/runtime checker; weights unchanged",
+  "cache_settings_update": "2026-09-25: document public server options and synthetic multi-turn tests; model unchanged"
 }

USAGE.md

@@ -100,6 +100,36 @@ encoder (`model.visual`). Their weights are retained and the release validation
 records their component execution. Speculative generation and integrated image/video
 chat are not implemented. Unsupported multimodal generation calls raise an error.
 
+## Server settings for limited memory
+
+The pinned MLX-LM server retains prompt caches for reuse by default. For a setup
+that favors lower memory use over reuse speed, stop the existing server and start
+it from the same patched Python environment with:
+
+```bash
+mlx_lm.server --model ukisai/Swift-1.5-4bit-MLX \
+  --host 127.0.0.1 --port 8080 \
+  --prompt-cache-size 0 \
+  --prompt-concurrency 1 --decode-concurrency 1 \
+  --prefill-step-size 512
+```
+
+This disables retained prompt caches, limits concurrency to one, and processes
+prefill in smaller chunks. The model weights and their quantization are unchanged.
+The active request still needs its own cache and temporary memory; these options
+do not set a total process-memory cap or guarantee that any context length fits.
+
+Disabling reuse can slow later turns because their history must be processed
+again. It does not remove history that the client includes in the next request.
+Start with a fresh short conversation and increase history while monitoring RAM.
+The model's configured context limit is distinct from the memory needed to run it.
+
+Validation scope: three successive HTTP requests, including a streaming request,
+passed on each small synthetic 4-bit and 5-bit architecture with these settings.
+Retained-cache accounting stayed at zero. These tests did not load the complete
+Swift checkpoint or establish full-model long-context capacity on any Mac.
+No macOS or MLX memory-limit overrides were used in those tests.
+
 Reproduce conversion only from the complete original Swift BF16 export identified
 in `QUANTIZATION_MANIFEST.json`, after verifying its 18 shards and original assets: