GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
veyra-ai
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
veyra-ai
Model Tree
veyra-ai/Veyra-30M-Base
Input Modalities
Output Modalities
Supported Functionality
veyra-ai/Veyra-30M-BaseVeyra-30M is a compact decoder-only transformer-style model.
This checkpoint was trained with masked ChatML supervised fine-tuning.
veyra-ai/veyra-30m-base-5b-tokenssft_masked_chatml_0100M.ptThe released checkpoint was selected from intermediate SFT milestones based on qualitative behavior, not only lowest training loss. Later SFT checkpoints reached lower loss but showed more generic refusal/template behavior, so checkpoint selection was based on output quality.
This is a 30M-parameter model and has significant limitations.
Known failure modes include:
This model is not fully safety-tuned. It may refuse some harmful requests, but refusal behavior is not reliable.
This model uses ChatML.
text
<|im_start|>systemYou are Veyra, a tiny local instruction model. Be concise, useful, casual, and lightly playful. Correct mistakes gently.<|im_end|><|im_start|>userWhat does a tokenizer do?<|im_end|><|im_start|>assistantpython
from transformers import AutoTokenizer, AutoModelForCausalLMimport torch model_id = "veyra-ai/veyra-30m-instruct" tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)model = AutoModelForCausalLM.from_pretrained(model_id, trust_remote_code=True) messages = [ { "role": "system", "content": "You are Veyra, a tiny local instruction model. Be concise, useful, casual, and lightly playful." }, { "role": "user", "content": "Explain what an API is using a simple analogy." },] prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)inputs = tokenizer(prompt, return_tensors="pt", add_special_tokens=False) with torch.no_grad(): outputs = model.generate( **inputs, max_new_tokens=120, temperature=0.3, top_k=40, do_sample=True, pad_token_id=tokenizer.pad_token_id, eos_token_id=[tokenizer.eos_token_id, tokenizer.convert_tokens_to_ids("<|im_end|>")], use_cache=True, ) print(tokenizer.decode(outputs[0], skip_special_tokens=False))The examples below are suggested prompts for trying the model. They are not benchmark results and should not be treated as representative guarantees.
text
Explain what an API is using a simple analogy.text
Return only JSON for a book with title='Dune', author='Frank Herbert', year=1965.text
Write a Python function that checks whether a number is even.text
What does FileNotFoundError usually mean in Python?text
Given these facts: color=blue, animal=otter, number=17. What animal was mentioned?Important tokenizer special tokens include:
text
<|bos|><|eos|><|pad|><|unk|><|im_start|><|im_end|><|tool_call|><|tool_result|><|context|><|reasoning|><|end_reasoning|><|answer|><|fim_prefix|><|fim_middle|><|fim_suffix|>For this checkpoint, standard ChatML with <|im_start|> and <|im_end|> is the recommended format.
If you use or build on this model, please retain attribution to Veyra AI.
Apache 2.0.
<|im_start|>systemYou are Veyra, a tiny local instruction model. Be concise, useful, casual, and lightly playful. Correct mistakes gently.<|im_end|><|im_start|>userWhat does a tokenizer do?<|im_end|><|im_start|>assistantfrom transformers import AutoTokenizer, AutoModelForCausalLMimport torch model_id = "veyra-ai/veyra-30m-instruct" tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)model = AutoModelForCausalLM.from_pretrained(model_id, trust_remote_code=True) messages = [ { "role": "system", "content": "You are Veyra, a tiny local instruction model. Be concise, useful, casual, and lightly playful." }, { "role": "user", "content": "Explain what an API is using a simple analogy." },] prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)inputs = tokenizer(prompt, return_tensors="pt", add_special_tokens=False) with torch.no_grad(): outputs = model.generate( **inputs, max_new_tokens=120, temperature=0.3, top_k=40, do_sample=True, pad_token_id=tokenizer.pad_token_id, eos_token_id=[tokenizer.eos_token_id, tokenizer.convert_tokens_to_ids("<|im_end|>")], use_cache=True, ) print(tokenizer.decode(outputs[0], skip_special_tokens=False))<|bos|><|eos|><|pad|><|unk|><|im_start|><|im_end|><|tool_call|><|tool_result|><|context|><|reasoning|><|end_reasoning|><|answer|><|fim_prefix|><|fim_middle|><|fim_suffix|>