minor fixes

README.md

@@ -77,6 +77,8 @@ UBATCH_SIZE=128
 PERF_PROFILE=balanced
 REASONING=False
 ENABLE_TOOLS=0
+CACHE_TYPE_K=q4_0
+CACHE_TYPE_V=q4_0
 ```
 
 Expected: faster first-token latency from prompt ingestion. Generation speed is still CPU-bound on the free tier, so a 2B Q4 model will usually remain in the low single-digit tokens/sec range.

start.sh

@@ -50,6 +50,21 @@ detect_memory_kb() {
   fi
 }
 
+normalize_arg_value() {
+  local value="$1"
+
+  value="${value//$'\r'/}"
+  value="${value//$'\n'/}"
+  value="${value#"${value%%[![:space:]]*}"}"
+  value="${value%"${value##*[![:space:]]}"}"
+
+  # Hugging Face env values are plain values. A copied shell-style "$q4_0"
+  # should be treated as "q4_0" instead of crashing llama-server.
+  value="${value#\$}"
+
+  printf '%s' "$value"
+}
+
 HOST_THREADS="$(nproc)"
 QUOTA_THREADS="$(detect_cpu_quota_threads)"
 if [ -z "${THREADS:-}" ]; then
@@ -147,8 +162,8 @@ else
 fi
 
 CTX_SIZE="${REQUESTED_CTX_SIZE:-$DEFAULT_CTX_SIZE}"
-CACHE_TYPE_K="${REQUESTED_CACHE_TYPE_K:-f16}"
-CACHE_TYPE_V="${REQUESTED_CACHE_TYPE_V:-f16}"
+CACHE_TYPE_K="$(normalize_arg_value "${REQUESTED_CACHE_TYPE_K:-f16}")"
+CACHE_TYPE_V="$(normalize_arg_value "${REQUESTED_CACHE_TYPE_V:-f16}")"
 PARALLEL=1
 
 echo "effective threads: $THREADS (host=$HOST_THREADS, quota=$QUOTA_THREADS)"