GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
drishtijs
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
drishtijs
Model Tree
Qwen/Qwen2.5-Coder-0.5B-Instruct
Input Modalities
Output Modalities
Supported Functionality
Install torch, transformers>=4.57.0 and safetensors in your environment.
The release smoke check used Transformers 4.57.0. Example:
python
from transformers import AutoModelForCausalLM, AutoTokenizer repo = "drishtijs/DrishX-1.0"revision = "v1.0.0"tokenizer = AutoTokenizer.from_pretrained(repo, revision=revision)model = AutoModelForCausalLM.from_pretrained( repo, revision=revision, dtype="auto", use_safetensors=True)messages = [{"role": "user", "content": "Explain a Drishti page route."}]prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)inputs = tokenizer(prompt, return_tensors="pt")output = model.generate(**inputs, max_new_tokens=128, do_sample=True, temperature=0.7, top_p=0.8, pad_token_id=tokenizer.eos_token_id)print(tokenizer.decode(output[0, inputs.input_ids.shape[1]:], skip_special_tokens=True))This release contains the existing local data/models/finetuned checkpoint,
not a new training run. The repository's scripts/train_model.py training recipe
uses the Qwen base model above and exports a full checkpoint. Exact run
hyperparameters and held-out quality metrics were not recorded with this checkpoint.
No benchmark or production-quality score is claimed.
validation.json records an offline CPU load, tokenizer, finite-logit and
deterministic generation smoke check using a synthetic greeting. This verifies
basic execution, not coding accuracy or comprehensive task performance.
This repository holds language-model weights. Filesystem edits, browser control, database retrieval, permission controls and project isolation belong to the DrishX application, not to the weights themselves. App installers and portable reference datasets remain in the DrishtiX dataset. The app's existing 7B GGUF is a separate model; this release does not replace it and does not include a GGUF conversion.
Generated code and advice can be incorrect and require review and testing.
Model versions use semantic versioning independently of app, npm and dataset
versions. Use v1.0.0 for reproducibility; main tracks the latest release.
release.json lists artifact sizes, SHA-256 hashes, architecture and source
revision. SHA256SUMS covers every published release file except itself.
No private chats, SQLite databases, credentials or raw training examples are uploaded.
from transformers import AutoModelForCausalLM, AutoTokenizer repo = "drishtijs/DrishX-1.0"revision = "v1.0.0"tokenizer = AutoTokenizer.from_pretrained(repo, revision=revision)model = AutoModelForCausalLM.from_pretrained( repo, revision=revision, dtype="auto", use_safetensors=True)messages = [{"role": "user", "content": "Explain a Drishti page route."}]prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)inputs = tokenizer(prompt, return_tensors="pt")output = model.generate(**inputs, max_new_tokens=128, do_sample=True, temperature=0.7, top_p=0.8, pad_token_id=tokenizer.eos_token_id)print(tokenizer.decode(output[0, inputs.input_ids.shape[1]:], skip_special_tokens=True))