GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
tangles2
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
tangles2
Model Tree
Qwen/Qwen2.5-0.5B-Instruct
Input Modalities
Output Modalities
Supported Functionality
The adapter trains the base model to call a sequence of finance tools in the correct order to resolve collections cases:
get_case / lookup_customer / search_invoices / search_payments / search_credit_memosupdate_case / create_exception / draft_slack_message / final_answerGiven a task description and the steps taken so far, the model outputs the next tool call as JSON:
json
{"tool": "search_invoices", "args": {"customer_id": "cus_globex", "invoice_id": "INV-2026-0413"}}| Parameter | Value |
|---|---|
| Base model | Qwen/Qwen2.5-0.5B-Instruct |
| LoRA rank | 8 |
| LoRA alpha | 16 |
| Target modules | q_proj, v_proj |
| Trainable params | 540,672 / 494,573,440 (0.11%) |
| Training examples | 28 (4 traces x ~7 steps each) |
| Epochs | 5 |
| Learning rate | 2e-4 (cosine schedule) |
| Hardware | Targon H200, bfloat16 |
| Training time | ~22 seconds |
Evaluated on 5 finance operations tasks across 8 scoring criteria.
| Agent | Strict Pass Rate | Avg Score | Broad Scans |
|---|---|---|---|
| Rule-based baseline | 0/5 (0%) | 86.3% | 2.0 |
| sklearn PolicyAgent | 4/5 (80%) | 90.0% | 0.0 |
| This model (LoRA) | 3/5 (60%) | 82.5% | 0.0 |
The model relies heavily on a PolicyAgent fallback (4-7 fallbacks per task) due to limited training data. With 100+ training examples the fallback rate would drop significantly.
task_credit_memo_reconciled fails across all agents due to only 1 training example of search_credit_memos.task_missing_evidence fails due to the model running past the correct stopping point (hits 16-step limit).python
from transformers import AutoModelForCausalLM, AutoTokenizerfrom peft import PeftModelimport torch base_model = "Qwen/Qwen2.5-0.5B-Instruct"adapter = "tangles2/acme-lora-qwen2.5-0.5b" tokenizer = AutoTokenizer.from_pretrained(adapter)model = AutoModelForCausalLM.from_pretrained(base_model, dtype=torch.bfloat16)model = PeftModel.from_pretrained(model, adapter)model.eval()See the benchmark repo for the full inference loop, mock environment, and scoring harness.
{"tool": "search_invoices", "args": {"customer_id": "cus_globex", "invoice_id": "INV-2026-0413"}}from transformers import AutoModelForCausalLM, AutoTokenizerfrom peft import PeftModelimport torch base_model = "Qwen/Qwen2.5-0.5B-Instruct"adapter = "tangles2/acme-lora-qwen2.5-0.5b" tokenizer = AutoTokenizer.from_pretrained(adapter)model = AutoModelForCausalLM.from_pretrained(base_model, dtype=torch.bfloat16)model = PeftModel.from_pretrained(model, adapter)model.eval()