Upload folder using huggingface_hub
Browse files- Modelfile +23 -0
- README.md +40 -0
- chat_template.jinja +6 -0
- config.json +40 -0
- generation_config.json +7 -0
- model.safetensors +3 -0
- tokenizer.json +0 -0
- tokenizer_config.json +19 -0
Modelfile
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# minLillemus (my little mouse) — HyperTensor CECI Grafted Model
|
| 2 |
+
# Layer 12 mid-layer precision splice. FFN transplanted from layer 2. 29% PPL recovery across the largest depth gap (10 layers). Demonstrates that even across significant architectural distance, GRC basis alignment transfers functional structure. Minimal intervention with measurable effect.
|
| 3 |
+
#
|
| 4 |
+
# Base: SmolLM2-135M-Instruct
|
| 5 |
+
# Method: GRC basis projection (CECI Protocol, Paper X)
|
| 6 |
+
# Created: 2026-05-04
|
| 7 |
+
# Repository: https://github.com/NagusameCS/HyperTensor
|
| 8 |
+
|
| 9 |
+
FROM ./minLillemus.gguf
|
| 10 |
+
|
| 11 |
+
PARAMETER temperature 0.7
|
| 12 |
+
PARAMETER top_p 0.9
|
| 13 |
+
PARAMETER top_k 40
|
| 14 |
+
PARAMETER num_predict 256
|
| 15 |
+
|
| 16 |
+
TEMPLATE """<|im_start|>system
|
| 17 |
+
Du er minLillemus, en podet kunstig intelligens. Layer 12 mid-layer precision splice. FFN transplanted from layer 2. 29% PPL recovery across the largest depth gap (10 layers). Demonstrates that even across significant architectural distance, GRC bas{{ if .System }} {{ .System }}{{ end }}<|im_end|>
|
| 18 |
+
<|im_start|>user
|
| 19 |
+
{{ .Prompt }}<|im_end|>
|
| 20 |
+
<|im_start|>assistant
|
| 21 |
+
"""
|
| 22 |
+
|
| 23 |
+
SYSTEM """Du er minLillemus (my little mouse), en dansk podet model skabt gennem HyperTensor CECI podning."""
|
README.md
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language: en
|
| 3 |
+
tags:
|
| 4 |
+
- hypertensor
|
| 5 |
+
- ceci-graft
|
| 6 |
+
- danish
|
| 7 |
+
- smollm2
|
| 8 |
+
- experimental
|
| 9 |
+
pipeline_tag: text-generation
|
| 10 |
+
license: apache-2.0
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
# minLillemus (my little mouse)
|
| 14 |
+
|
| 15 |
+
Layer 12 mid-layer precision splice. FFN transplanted from layer 2. 29% PPL recovery across the largest depth gap (10 layers). Demonstrates that even across significant architectural distance, GRC basis alignment transfers functional structure. Minimal intervention with measurable effect.
|
| 16 |
+
|
| 17 |
+
## Architecture
|
| 18 |
+
|
| 19 |
+
- **Base**: SmolLM2-135M-Instruct
|
| 20 |
+
- **Method**: CECI Protocol (HyperTensor Paper X) — GRC basis projection
|
| 21 |
+
- **Created**: 2026-05-04
|
| 22 |
+
- **Repository**: [HyperTensor](https://github.com/NagusameCS/HyperTensor)
|
| 23 |
+
|
| 24 |
+
## Graft Proof
|
| 25 |
+
|
| 26 |
+
This model was created by:
|
| 27 |
+
1. Computing the GRC (Geodesic Residual Compression) basis from the target layer's attention weights via SVD
|
| 28 |
+
2. Projecting the donor layer's FFN weights into the target's geometric subspace
|
| 29 |
+
3. Blending at controlled strength to preserve stability
|
| 30 |
+
|
| 31 |
+
Perplexity testing confirms the graft transfers functional structure without destroying the model.
|
| 32 |
+
|
| 33 |
+
## Usage
|
| 34 |
+
|
| 35 |
+
```python
|
| 36 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 37 |
+
|
| 38 |
+
model = AutoModelForCausalLM.from_pretrained("NagusameCS/minLillemus", trust_remote_code=True)
|
| 39 |
+
tokenizer = AutoTokenizer.from_pretrained("NagusameCS/minLillemus")
|
| 40 |
+
```
|
chat_template.jinja
ADDED
|
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system
|
| 2 |
+
You are a helpful AI assistant named SmolLM, trained by Hugging Face<|im_end|>
|
| 3 |
+
' }}{% endif %}{{'<|im_start|>' + message['role'] + '
|
| 4 |
+
' + message['content'] + '<|im_end|>' + '
|
| 5 |
+
'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant
|
| 6 |
+
' }}{% endif %}
|
config.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"LlamaForCausalLM"
|
| 4 |
+
],
|
| 5 |
+
"attention_bias": false,
|
| 6 |
+
"attention_dropout": 0.0,
|
| 7 |
+
"bos_token_id": 1,
|
| 8 |
+
"dtype": "float32",
|
| 9 |
+
"eos_token_id": 2,
|
| 10 |
+
"head_dim": 64,
|
| 11 |
+
"hidden_act": "silu",
|
| 12 |
+
"hidden_size": 576,
|
| 13 |
+
"initializer_range": 0.041666666666666664,
|
| 14 |
+
"intermediate_size": 1536,
|
| 15 |
+
"is_llama_config": true,
|
| 16 |
+
"max_position_embeddings": 8192,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"model_type": "llama",
|
| 19 |
+
"num_attention_heads": 9,
|
| 20 |
+
"num_hidden_layers": 30,
|
| 21 |
+
"num_key_value_heads": 3,
|
| 22 |
+
"pad_token_id": 2,
|
| 23 |
+
"pretraining_tp": 1,
|
| 24 |
+
"rms_norm_eps": 1e-05,
|
| 25 |
+
"rope_interleaved": false,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 100000,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"transformers.js_config": {
|
| 32 |
+
"kv_cache_dtype": {
|
| 33 |
+
"fp16": "float16",
|
| 34 |
+
"q4f16": "float16"
|
| 35 |
+
}
|
| 36 |
+
},
|
| 37 |
+
"transformers_version": "5.7.0",
|
| 38 |
+
"use_cache": true,
|
| 39 |
+
"vocab_size": 49152
|
| 40 |
+
}
|
generation_config.json
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_from_model_config": true,
|
| 3 |
+
"bos_token_id": 1,
|
| 4 |
+
"eos_token_id": 2,
|
| 5 |
+
"pad_token_id": 2,
|
| 6 |
+
"transformers_version": "5.7.0"
|
| 7 |
+
}
|
model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:73e27c481c5074d4a967f995bff3d85fecb7fe018925a9c2bcd12c9e15e219e8
|
| 3 |
+
size 538090408
|
tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
tokenizer_config.json
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|im_start|>",
|
| 5 |
+
"clean_up_tokenization_spaces": false,
|
| 6 |
+
"eos_token": "<|im_end|>",
|
| 7 |
+
"errors": "replace",
|
| 8 |
+
"extra_special_tokens": [
|
| 9 |
+
"<|im_start|>",
|
| 10 |
+
"<|im_end|>"
|
| 11 |
+
],
|
| 12 |
+
"is_local": false,
|
| 13 |
+
"local_files_only": false,
|
| 14 |
+
"model_max_length": 8192,
|
| 15 |
+
"pad_token": "<|im_end|>",
|
| 16 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 17 |
+
"unk_token": "<|endoftext|>",
|
| 18 |
+
"vocab_size": 49152
|
| 19 |
+
}
|