# ============================================================================
# Ollama Modelfile for GLM-5.2 (GGUF version)
# ============================================================================
# PREREQUISITE: Mount the GGUF repo first:
#   hf-mount start --hf-token $HF_TOKEN repo unsloth/GLM-5.2-GGUF /mnt/glm52-gguf
#
# Then adjust the FROM path below to match the actual GGUF filename.
# Available quants from unsloth/GLM-5.2-GGUF typically include:
#   Q2_K, Q3_K_M, Q4_K_M, Q5_K_M, Q6_K, Q8_0, F16
#
# Choose based on your available RAM/VRAM:
#   Q4_K_M ≈ good balance of quality and size
#   Q2_K   ≈ smallest, lower quality
#   Q8_0   ≈ near-lossless, largest
# ============================================================================

FROM /mnt/glm52-gguf/GLM-5.2-Q4_K_M.gguf

# --- Generation Parameters ---
PARAMETER temperature 0.7
PARAMETER top_p 0.9
PARAMETER top_k 50
PARAMETER num_ctx 32768
PARAMETER repeat_penalty 1.1
PARAMETER num_predict -1
PARAMETER seed 42

# --- Stop Tokens (GLM format) ---
PARAMETER stop "<|endoftext|>"
PARAMETER stop "<|user|>"
PARAMETER stop "<|assistant|>"
PARAMETER stop "<|observation|>"

# --- System Prompt ---
SYSTEM """You are GLM-5.2, a large language model developed by Zhipu AI. You are helpful, harmless, and honest. Respond clearly and concisely."""

# --- Chat Template (GLM conversation format) ---
TEMPLATE """{{ if .System }}<|system|>
{{ .System }}<|endoftext|>
{{ end }}{{ if .Prompt }}<|user|>
{{ .Prompt }}<|endoftext|>
{{ end }}<|assistant|>
{{ .Response }}<|endoftext|>
"""

# --- License ---
LICENSE "MIT"

# ============================================================================
# BUILD COMMAND:
#   ollama create glm-5.2 -f Modelfile
#
# RUN COMMAND:
#   ollama run glm-5.2
#
# The ollama create command will:
#   1. Read the GGUF from the mounted path (lazy I/O via hf-mount)
#   2. Copy the weight blob into ~/.ollama/models/blobs/
#   3. Auto-generate a manifest at:
#      ~/.ollama/models/manifests/registry.ollama.ai/library/glm-5.2/latest
#
# NOTE: `ollama create` copies the GGUF into Ollama's blob store.
# This means the full GGUF file WILL be copied locally during create.
# To avoid this, use vLLM instead — it reads directly from the mount.
# ============================================================================