GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
MallikarjunJadi
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
MallikarjunJadi
Model Tree
Qwen/Qwen2.5-1.5B-Instruct
Input Modalities
Output Modalities
Supported Functionality
| LoRA | QLoRA | |
|---|---|---|
| rank / alpha / dropout | 16 / 32 / 0.05 | 16 / 32 / 0.05 |
| target modules | q,k,v,o_proj | q,k,v,o_proj |
| base precision | bf16 | 4-bit NF4 (double quant), bf16 compute |
| micro-batch x grad-accum | 4 x 4 | 2 x 8 |
| learning rate | 2e-4 | 1e-4 |
| epochs | 3 | 3 |
LoRA and QLoRA differ in learning rate / micro-batch, so their comparison is indicative, not perfectly controlled.
| model | rouge_l | bertscore_f1 | token_f1 | ref_contained | exact_match | avg_words_pred | avg_words_ref |
|---|---|---|---|---|---|---|---|
| base | 0.1074 | 0.8482 | 0.1590 | 0.0350 | 0.0000 | 94.6700 | 23.6200 |
| lora | 0.3559 | 0.9046 | 0.3973 |
rouge_l/bertscore_f1: overlap/semantic similarity with short reference answers; token_f1: token-level F1;
ref_contained: reference answer appears inside the prediction. Automatic metrics are proxies, not agronomic validation.
python
from agri_llm import AgriLLMllm = AgriLLM(adapters={"qlora": "<this-repo-or-folder>"})print(llm.generate("How do I control aphids on cotton?", mode="qlora"))Trained on community-curated Q&A, not agronomist-verified. No retrieval, so it can state wrong facts confidently (see failure_candidates.csv). Research prototype - not a substitute for professional agricultural advice.
| 0.1150 |
| 0.1050 |
| 24.1550 |
| 23.6200 |
| qlora | 0.2189 | 0.8784 | 0.2649 | 0.0350 | 0.0200 | 35.3800 | 23.6200 |