GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
artyomboyko
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
artyomboyko
Model Tree
Qwen/Qwen3.5-2B-Base
Input Modalities
Output Modalities
Supported Functionality
| Property | Value |
|---|---|
| Base model | Qwen/Qwen3.5-2B-Base |
| Method | Prompt Tuning |
| Dataset | stanfordnlp/sst2 |
| Task | Sentiment classification |
| Labels | negative, positive |
| Virtual tokens | 16 |
| Prompt initialization | Classify the sentiment of the movie review as positive or negative. |
| Adapter size | 0.13 MiB |
Evaluation scope: full SST-2 validation split.
| Metric | Base model | Prompt Tuning |
|---|---|---|
| Generation accuracy | 3.10% | 94.27% |
| Forced-choice accuracy | 51.49% | 94.27% |
| Generation Macro F1 | 0.0573 | 0.9427 |
| Forced-choice Macro F1 | 0.3502 | 0.9427 |
| Perplexity | — | 1.0869 |
python
import torchfrom peft import PeftModelfrom transformers import (AutoModelForCausalLM,AutoTokenizer,)BASE_MODEL_ID = "Qwen/Qwen3.5-2B-Base"ADAPTER_ID = "artyomboyko/qwen3.5-2b-sst2-prompt-tuning"tokenizer = (AutoTokenizer.from_pretrained(BASE_MODEL_ID))if tokenizer.pad_token_id is None:tokenizer.pad_token = (tokenizer.eos_token)base_model = (AutoModelForCausalLM.from_pretrained(BASE_MODEL_ID,dtype="auto",))model = PeftModel.from_pretrained(base_model,ADAPTER_ID,)device = torch.device("cuda"if torch.cuda.is_available()else "cpu")model = model.to(device)model.eval()review = ("a wonderfully acted ""and moving story")prompt = ("Classify the sentiment of this movie review as positive or negative.\n"f"Review: {review}\n""Sentiment:")inputs = tokenizer(prompt,return_tensors="pt",).to(device)with torch.inference_mode():outputs = model.generate(**inputs,max_new_tokens=4,do_sample=False,pad_token_id=(tokenizer.eos_token_id),)generated = outputs[:,inputs["input_ids"].shape[1]:,]prediction = tokenizer.decode(generated[0],skip_special_tokens=True,).strip()print(prediction)
Qwen/Qwen3.5-2B-Base.