Fix Modelfile: add RENDERER/PARSER qwen3.5 for tool calling; relative GGUF path; remove hardcoded TEMPLATE override

Modelfile

@@ -1,12 +1,39 @@
-FROM /home/roko/Documents/Projects/Adjacent/distill_qwen/outputs/gguf/qwen3.5-9b-qwen3.6-reasoning-distilled/qwen3.5-9b-qwen3.6-reasoning-distilled.q4km.gguf
-TEMPLATE """{{- if .System }}<|im_start|>system
-{{ .System }}<|im_end|>
-{{ end }}{{- range .Messages }}<|im_start|>{{ .Role }}
-{{ .Content }}<|im_end|>
-{{ end }}<|im_start|>assistant
-"""
-PARAMETER stop "<|im_end|>"
-PARAMETER stop "<|im_start|>"
+# Modelfile for Ollama — qwen3.5-9b-qwen3.6-reasoning-distilled
+#
+# CRITICAL: `RENDERER qwen3.5` and `PARSER qwen3.5` are REQUIRED for tool
+# calling / function calling to work. Without them Ollama falls back to
+# generic text handling, `ollama show` reports "does not support tools",
+# and /api/chat ignores the `tools` field.
+#
+# The embedded chat template in the GGUF already supports tool use. These
+# two directives wire Ollama's native Go renderer + parser for Qwen3.5 so
+# the model's <tool_call> output gets translated to structured
+# message.tool_calls in the API response.
+#
+# Default here uses the Q4_K_M (~5.6 GB) quant. To use Q8_0 (~9.5 GB)
+# replace the FROM line with:
+#   FROM ./qwen3.5-9b-qwen3.6-reasoning-distilled.q80.gguf
+# To use F16 (~17.9 GB):
+#   FROM ./qwen3.5-9b-qwen3.6-reasoning-distilled.f16.gguf
+#
+# To create the model locally after downloading the GGUF + this Modelfile:
+#   ollama create qwen3.5-9b-qwen3.6-distilled:q4km -f Modelfile
+#   ollama show   qwen3.5-9b-qwen3.6-distilled:q4km   # should list: tools, thinking
+#   ollama run    qwen3.5-9b-qwen3.6-distilled:q4km
+#
+# Or pull directly from this HF repo (specify the quant as the tag):
+#   ollama pull hf.co/cudabenchmarktest/qwen3.5-9b-qwen3.6-reasoning-distilled-GGUF:Q4_K_M
+#   ollama pull hf.co/cudabenchmarktest/qwen3.5-9b-qwen3.6-reasoning-distilled-GGUF:Q8_0
+#   ollama pull hf.co/cudabenchmarktest/qwen3.5-9b-qwen3.6-reasoning-distilled-GGUF:F16
+# NOTE: Direct `hf.co/...` pulls do not apply this Modelfile — Ollama will
+# auto-derive one from GGUF metadata. If the auto-derived model loses tool
+# support, download this Modelfile + the GGUF and `ollama create` manually.
+
+FROM ./qwen3.5-9b-qwen3.6-reasoning-distilled.q4km.gguf
+
+RENDERER qwen3.5
+PARSER qwen3.5
+
 PARAMETER temperature 0.6
 PARAMETER top_p 0.95
-SYSTEM """You are a reasoning model distilled from Qwen3.6-plus. Think step-by-step inside <think></think> tags, then provide a clear final answer."""
+PARAMETER stop "<|im_end|>"