File size: 2,512 Bytes
cc166af ba76308 1752318 ba76308 cc166af ba76308 81ed35c 9e5df5f ba76308 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 | ---
library_name: transformers
pipeline_tag: text-generation
license: apache-2.0
language:
- en
tags:
- text-generation
- conversational
- qwen2
- from-scratch
- english
---
# Tensor 1.5
Tensor 1.5 is a compact conversational language model built from scratch by **VGribAI**.
It opens the Tensor line: the first model whose name carries no parameter size, the first one
trained exclusively in English, and a complete assistant rather than a base checkpoint, since it
has gone through supervised fine-tuning.
## What makes it different
- **No size in the name.** Tensor 1.5 starts a naming scheme where the version identifies the
model, not the parameter count.
- **English only.** The tokenizer, the pretraining and the fine-tuning are all English. The model
is not expected to follow other languages.
- **A finished assistant.** Tensor 1.5 follows the chat format, keeps the assistant role and stops
on its own turn end. It is a model you talk to, not a text continuation base.
## Architecture
| | |
|---|---|
| Decoder | Qwen2 style (12 layers, hidden 640, GQA 8 query / 2 KV heads, SwiGLU) |
| Parameters | 50.26M, tied embeddings |
| Vocabulary | 4000 byte-level BPE, trained for this model |
| Context | 512 tokens |
| Default answer length | 64 tokens |
| Weights | fp32 safetensors, plus a f16 GGUF build |
## Intended use
Short answers, quick summaries, simple rewrites and light chit-chat on weak hardware: CPUs,
old laptops, single-board computers and phones.
## Usage
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("tensor-1.5")
model = AutoModelForCausalLM.from_pretrained("tensor-1.5")
msgs = [{"role": "user", "content": "Who are you?"}]
ids = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt")
out = model.generate(ids, max_new_tokens=64, eos_token_id=[0, 2], pad_token_id=2)
print(tok.decode(out[0, ids.shape[1]:], skip_special_tokens=True))
```
## Chat format
```
<|im_start|>system
You are Tensor, a small assistant made by VGribAI.<|im_end|>
<|im_start|>user
Hello!<|im_end|>
<|im_start|>assistant
Hello! How can I help you?<|im_end|>
```
Stop tokens: `<|im_end|>` (id 2) and `<|eos|>` (id 0).
## Limits
- English only.
- 512 tokens of context, 64 tokens per answer.
- A very small model: fluent phrasing, but no reliable factual knowledge, arithmetic or long
reasoning. Treat answers as text, not as facts.
- No memory beyond the current context window. |