GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
hotdogs
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
hotdogs
Model Tree
Qwen/Qwen3.8-27B
Input Modalities
Output Modalities
Supported Functionality
| metric | value |
|---|---|
| base model | Qwen/Qwen3.8-27B |
| source model | Cloudflare/clef |
| tensor pairs | 186 |
| rank | 64 |
| energy retained (mean) | 0.8926 |
| energy retained (min) | 0.6796 |
| energy retained (max) | 1.0000 |
| reconstruction rel. err (mean) | 0.2814 |
Only 186 of 496 linear layers show a significant weight delta (≥0.001 norm), indicating a lighter fine-tune. The retained delta is highly low-rank (mean energy ≈0.89, reconstruction error ≈0.28).
adapter_model.safetensors — full PEFT/LoRA adapter (186 pairs, float32, ~657 MB)adapter_config.json — PEFT config (target modules auto-derived)extraction_stats.json — per-tensor SVD metricsclef_qwen3.8-27B_svd_lora.gguf — llama.cpp GGUF LoRA (300 tensors, ~653 MB)bash
llama-cli \ -m /path/to/Qwen3.8-27B.gguf \ --lo-ra clef_qwen3.8-27B_svd_lora.gguf 1.0 \ -p "Hello!"python
from transformers import AutoModelForCausalLMfrom peft import PeftModel base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.8-27B", torch_dtype="auto")model = PeftModel.from_pretrained(base, "hotdogs/qwen3.8-27B-clef_svd_lora")The GGUF contains 300 tensors. in_proj_a / in_proj_b (GDN gates) are
excluded because convert_lora_to_gguf.py does not map those GDN gate
names; 1-D norm weights and 3-D conv1d are not LoRA-able and are omitted. The
full PEFT adapter retains all 186 pairs.
Generated by exact truncated SVD on the weight difference
Cloudflare/clef − Qwen3.8-27B. Target keys aligned directly with the
base model key space (no remap needed).
from transformers import AutoModelForCausalLMfrom peft import PeftModel base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.8-27B", torch_dtype="auto")model = PeftModel.from_pretrained(base, "hotdogs/qwen3.8-27B-clef_svd_lora")llama-cli \ -m /path/to/Qwen3.8-27B.gguf \ --lo-ra clef_qwen3.8-27B_svd_lora.gguf 1.0 \ -p "Hello!"