初始化项目，由ModelHub XC社区提供模型

Model: xxrickyxx/Ailo152m-v2
Source: Original Platform

.gitattributes

@@ -0,0 +1,3 @@
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.gguf filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,251 @@
+---
+license: cc-by-nc-sa-4.0
+language:
+- en
+tags:
+- text-generation
+- transformer
+- instruction-tuned
+- reasoning
+- web-search
+- rag
+- small-language-model
+- edge-ai
+- on-device
+- cpu-inference
+- ollama
+- gguf
+- causal-lm
+- conversational
+pipeline_tag: text-generation
+library_name: gguf
+model-index:
+- name: AILO-152M-v2
+  results: []
+---
+
+# AILO-152M-v2  Tiny LLM with Chat, Reasoning & Web Search ⚡
+
+> **A 152M-parameter language model that runs on almost anything**  laptops, old PCs, even a Raspberry Pi  yet does instruction-following chat, step-by-step reasoning, and **web search** for fresh facts.
+
+**AILO** (Artificial Intelligence Language Operator) is a compact, fast, from-scratch transformer. v2 turns the original base model into a real assistant: it answers questions, thinks before answering, and can use **live web results** to answer about things it was never trained on.
+
+```bash
+ollama run Alieno/ailo-152m-v2
+```
+
+| | |
+|---|---|
+| 🧠 **Parameters** | 151.9M |
+| ⚡ **Speed** | up to **384 tok/s** (GPU), runs on **CPU & edge** |
+| 📦 **Size** | 97 MB (q4_k_m) – 305 MB (f16) |
+| 🌐 **Web search** | yes (context-following) |
+| 💭 **Reasoning** | yes (`<think>`) |
+| 🪶 **Min RAM** | ~300 MB |
+
+---
+
+## ✨ Why AILO-152M-v2?
+
+- **Runs anywhere**  97 MB quantized, ~300 MB RAM. Old laptops, mini-PCs, Raspberry Pi, phones.
+- **Fast**  fastest in its class (see benchmarks). Real-time chat even on modest hardware.
+- **Web-aware**  trained for *context-following*, so it answers from fresh search results instead of stale memory.
+- **Distilled from a bigger model**  answers learned from **Gemma 3 4B** (knowledge distillation): richer, better-structured replies than its size suggests.
+- **Honest small model**  strong at concise factual Q&A and conversation; pairs with tools for exact math.
+- **Open & local**  no cloud, full privacy, drop-in for Ollama.
+
+**Great for:** edge/on-device AI, offline assistants, learning how LLMs work, fast prototyping, low-power servers, privacy-first chatbots.
+
+---
+
+## 🚀 Quick start
+
+### Ollama (recommended)
+```bash
+ollama run Alieno/ailo-152m-v2
+>>> What is the capital of Italy?
+The capital city of Italy is Rome.
+```
+
+Tags: `:latest` / `:q8_0` (best quality, 156 MB) · `:q4_k_m` (smallest, 97 MB) · `:f16` (full precision, 305 MB)
+
+### API
+```bash
+curl http://localhost:11434/api/chat -d '{
+  "model": "Alieno/ailo-152m-v2",
+  "messages": [{"role": "user", "content": "Explain what gravity is."}]
+}'
+```
+
+---
+
+## 🏆 Benchmarks
+
+Evaluated via Ollama `/api/chat` on factual QA, reasoning and coherence vs comparable and **larger** models:
+
+| Model | Params | Factual | Reasoning | Coherence | Speed (tok/s) |
+|-------|--------|---------|-----------|-----------|---------------|
+| **AILO-152M-v2** | **152M** | **7/8** | 1–2/5 | **100%** | **384** 🥇 |
+| SmolLM2 | 135M | 8/8 | 1/5 | 98% | 403 |
+| Qwen2.5 | 500M | 8/8 | 3–4/5 | 96% | 213 |
+| TinyLlama | 1.1B | 8/8 | 1–2/5 | 97% | 260 |
+
+- 🥇 **Top coherence** (100%  virtually no repetition) and among the **fastest**.
+- Competitive on factual accuracy with models its size and **larger**.
+- Trails only bigger instruction-tuned models on multi-step reasoning  expected for the smallest, from-scratch model here.
+
+> Measured on an NVIDIA RTX 5060 Ti. Reasoning has run-to-run variance on an 8/5-question micro-suite.
+
+---
+
+## 🖥️ Hardware & performance
+
+AILO-152M is tiny, so it runs **far beyond high-end GPUs**  including old and low-power hardware. Approximate generation speed (q4_k_m, ~97 MB):
+
+| Hardware | Type | Est. speed (tok/s) | Notes |
+|----------|------|--------------------|-------|
+| RTX 5060 Ti / 4070+ | Modern GPU | **350–450** | ✅ measured: 384 (q8_0) |
+| RTX 3060 / 2070 | Mid GPU | ~250–350 | smooth real-time |
+| GTX 1660 / 1060 | **Older GPU** | ~150–220 | still real-time |
+| GTX 1050 / MX150 | **Old laptop GPU** | ~90–140 | very usable |
+| Ryzen 7 / Core i7 (recent) | Modern CPU | ~45–80 | no GPU needed |
+| Core i5 ~2015 | **Old CPU** | ~18–30 | usable for chat |
+| Raspberry Pi 5 | **SBC / edge** | ~10–16 | runs offline |
+| Raspberry Pi 4 | **Low-power SBC** | ~5–9 | runs offline |
+| Recent smartphone | **Mobile** | ~15–35 | via llama.cpp/Termux |
+
+*Estimates except the measured RTX 5060 Ti; real numbers vary with quantization, RAM bandwidth and build flags. The takeaway: **AILO runs even where larger models can't load at all.***
+
+**Minimum requirements:** ~300 MB RAM (q4_k_m), any x86-64 / ARM CPU. No GPU required.
+
+---
+
+## 💬 Chat format
+
+Trained on this template (tags are plain GPT-2 BPE sequences  no vocab extension):
+```
+<|user|>
+{question}
+<|assistant|>
+<think>{optional reasoning}</think>
+{answer}<|end|>
+```
+
+---
+
+## 🌐 Web search (fresh facts)
+
+AILO v2 is trained for **context-following with override**: give it search results and it answers from them  **even when they contradict its training-time knowledge**, so it can use *up-to-date* facts. When no context is given, it falls back to its own (true) knowledge.
+
+A ready pipeline is included (`ailo_web.py`): DuckDuckGo → instant-answer + **semantic re-ranking** (MiniLM) with language/relevance filters → short clean context (fits the 512-token window) → AILO answers.
+
+```bash
+python ailo_web.py "What is the tallest mountain in the world?"
+# -> "Mount Everest, at 8,848 meters."
+```
+
+*This is how a 152M model can answer about events it never saw in training.*
+
+---
+
+## 💭 Reasoning (thinking)
+
+The model declares the **`thinking` capability**: set `"think": true` and the reasoning trace is returned in `message.thinking`, separate from the answer (shown in a dedicated box in the Ollama desktop app). Best on reasoning-style prompts; for exact math, pair with a calculator tool.
+
+---
+
+## 🐍 Python (Transformers)
+```python
+from huggingface_hub import hf_hub_download
+import torch, tiktoken, sys
+repo = "xxrickyxx/ailo-152m-v2"
+for f in ["config.json","configuration_ailo.py","modeling_ailo.py","pytorch_model.bin"]:
+    hf_hub_download(repo_id=repo, filename=f, local_dir="ailo_v2")
+sys.path.insert(0, "ailo_v2")
+from modeling_ailo import AILOForCausalLM
+from configuration_ailo import AILOConfig
+model = AILOForCausalLM(AILOConfig())
+model.load_state_dict(torch.load("ailo_v2/pytorch_model.bin", map_location="cpu"), strict=False)
+model.eval()
+tok = tiktoken.get_encoding("gpt2")
+ids = torch.tensor([tok.encode_ordinary("<|user|>\nWhat is the capital of Italy?\n<|assistant|>\n")])
+print(tok.decode(model.generate(ids, max_new_tokens=40, temperature=0.3)[0].tolist()))
+```
+
+---
+
+## 📐 Model details
+
+| Property | Value |
+|----------|-------|
+| Parameters | 151.9M |
+| Architecture | Decoder-only Transformer (LayerNorm · RoPE · SwiGLU) |
+| Layers / Hidden / Heads | 12 / 768 / 12 |
+| Context length | 512 tokens |
+| Vocabulary | 50,257 (GPT-2 BPE) |
+| Base | AILO-152M (FineWeb-Edu, 182k steps) |
+| Fine-tuning | SFT + **distillation from Gemma 3 4B**: instruction + reasoning (GSM8K) + context-following (SQuAD) + context-override + tool-use |
+| Formats | GGUF (q4_k_m, q8_0, f16) + PyTorch |
+
+---
+
+## ⚠️ Limitations
+- 152M params: limited world knowledge and multi-step reasoning vs larger models.
+- 512-token context: best with short, focused prompts; not for long documents.
+- Web-search quality depends on search-result quality; best for well-defined factual questions.
+- For exact arithmetic, use the tool/agent layer (the calculator does the math).
+- Primarily English.
+
+---
+
+## 📜 License
+
+This project uses a **dual-license** model.
+
+### 🆓 Non-Commercial License
+Released under **Creative Commons Attribution-NonCommercial-ShareAlike 4.0 (CC BY-NC-SA 4.0)**.
+
+You are free to:
+- Use the model for **research, education, and personal projects**
+- **Modify and fine-tune** the model
+- **Redistribute derivatives** under the same license
+
+You must:
+- Provide **attribution**
+- Keep the **same license** for derivative works
+- **Not** use the model for **commercial purposes**
+
+### 💼 Commercial License
+Commercial use of AILO-152M is **not permitted** under the free license. Commercial use includes:
+- Integration into paid products or services
+- Use in SaaS platforms, APIs, or enterprise systems
+- Any application that generates revenue directly or indirectly
+
+For commercial licensing, a separate paid agreement (royalty or license fee) is required. Please contact the author.
+
+---
+
+## 📬 Contact
+
+For research collaboration or commercial licensing inquiries, contact the project maintainer:
+
+**Riccardo Sparacino**  [LinkedIn](https://www.linkedin.com/in/riccardo-sparacino-developer-php-javascript-mysql-app-ios-android/)
+
+---
+
+## 📑 Citation
+```bibtex
+@misc{ailo152m_v2_2026,
+  title  = {AILO-152M-v2: A Tiny Instruction-Tuned LLM with Reasoning and Web Search},
+  author = {Sparacino, Riccardo},
+  year   = {2026},
+  note   = {Dual-licensed CC BY-NC-SA 4.0 / commercial}
+}
+```
+
+## 🙏 Acknowledgments
+Built with [Ollama](https://ollama.com) and [llama.cpp](https://github.com/ggerganov/llama.cpp). Fine-tuning data: Alpaca-cleaned, GSM8K, SQuAD. Knowledge-distillation teacher: **Gemma 3 4B**. Embeddings for web re-ranking: sentence-transformers MiniLM.
+
+---
+
+*Keywords: small language model, tiny LLM, 152M, efficient LLM, edge AI, on-device LLM, CPU inference, Raspberry Pi LLM, Ollama model, GGUF, instruction-tuned, reasoning model, web search LLM, RAG, offline assistant, low-resource, fast inference.*

ailo-152m-v2-f16.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a3fd5dc78a687b98b822a714dd2d593513819b0d08d4f8531a3afbe3bc21d8b
+size 305537696

ailo-152m-v2-q4_k_m.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4413fc146577c6756747e885f55703773d9baaf6fb7708d2b2e95b11f20d273
+size 101775424

ailo-152m-v2-q8_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6cec6db0bf707d5a62263c570a24cfc7ff4a1dfe7465020966078f776fa84833
+size 163184416

config.json

@@ -0,0 +1,23 @@
+{
+  "architectures": [
+    "AILOForCausalLM"
+  ],
+  "model_type": "ailo",
+  "vocab_size": 50257,
+  "hidden_size": 768,
+  "num_hidden_layers": 12,
+  "num_attention_heads": 12,
+  "intermediate_size": 3072,
+  "max_position_embeddings": 512,
+  "hidden_dropout_prob": 0.1,
+  "attention_probs_dropout_prob": 0.1,
+  "bos_token_id": 50256,
+  "eos_token_id": 50256,
+  "pad_token_id": 50256,
+  "torch_dtype": "float32",
+  "transformers_version": "4.57.0",
+  "auto_map": {
+    "AutoConfig": "configuration_ailo.AILOConfig",
+    "AutoModelForCausalLM": "modeling_ailo.AILOForCausalLM"
+  }
+}

configuration_ailo.py

@@ -0,0 +1,41 @@
+"""
+AILO Configuration for HuggingFace Transformers
+"""
+
+from transformers import PretrainedConfig
+
+
+class AILOConfig(PretrainedConfig):
+    """Configuration class for AILO model."""
+    
+    model_type = "ailo"
+    
+    def __init__(
+        self,
+        vocab_size: int = 50257,
+        hidden_size: int = 768,
+        num_hidden_layers: int = 12,
+        num_attention_heads: int = 12,
+        intermediate_size: int = 3072,
+        max_position_embeddings: int = 512,
+        hidden_dropout_prob: float = 0.1,
+        attention_probs_dropout_prob: float = 0.1,
+        bos_token_id: int = 50256,
+        eos_token_id: int = 50256,
+        pad_token_id: int = 50256,
+        **kwargs
+    ):
+        super().__init__(
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            pad_token_id=pad_token_id,
+            **kwargs
+        )
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.intermediate_size = intermediate_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_dropout_prob = hidden_dropout_prob
+        self.attention_probs_dropout_prob = attention_probs_dropout_prob

modeling_ailo.py

@@ -0,0 +1,216 @@
+"""
+AILO Model for HuggingFace Transformers - Matching original architecture
+"""
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import math
+from typing import Optional, Tuple, Union
+
+from transformers import PreTrainedModel
+from transformers.generation import GenerationMixin
+from transformers.modeling_outputs import CausalLMOutputWithPast
+
+try:
+    from .configuration_ailo import AILOConfig
+except ImportError:
+    from configuration_ailo import AILOConfig
+
+
+class RotaryPositionalEmbedding(nn.Module):
+    """Rotary Position Embedding (RoPE)."""
+    
+    def __init__(self, dim: int, max_seq_len: int = 512, base: int = 10000):
+        super().__init__()
+        inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2).float() / dim))
+        self.register_buffer("inv_freq", inv_freq)
+        self.max_seq_len = max_seq_len
+        
+    def forward(self, x: torch.Tensor, seq_len: int) -> Tuple[torch.Tensor, torch.Tensor]:
+        t = torch.arange(seq_len, device=x.device).type_as(self.inv_freq)
+        freqs = torch.einsum("i,j->ij", t, self.inv_freq)
+        emb = torch.cat((freqs, freqs), dim=-1)
+        return emb.cos(), emb.sin()
+
+
+def apply_rotary_pos_emb(q, k, cos, sin):
+    """Apply rotary position embedding."""
+    def rotate_half(x):
+        x1, x2 = x[..., :x.shape[-1]//2], x[..., x.shape[-1]//2:]
+        return torch.cat((-x2, x1), dim=-1)
+    
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+class AILOAttention(nn.Module):
+    """Multi-head attention matching original structure."""
+    
+    def __init__(self, config: AILOConfig):
+        super().__init__()
+        self.n_heads = config.num_attention_heads
+        self.head_dim = config.hidden_size // config.num_attention_heads
+        self.scale = self.head_dim ** -0.5
+        
+        # Match original: separate q, k, v projections
+        self.q_proj = nn.Linear(config.hidden_size, config.hidden_size, bias=False)
+        self.k_proj = nn.Linear(config.hidden_size, config.hidden_size, bias=False)
+        self.v_proj = nn.Linear(config.hidden_size, config.hidden_size, bias=False)
+        self.out_proj = nn.Linear(config.hidden_size, config.hidden_size, bias=False)
+        
+        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
+        self.rotary = RotaryPositionalEmbedding(self.head_dim, config.max_position_embeddings)
+        
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        B, T, C = x.shape
+        
+        q = self.q_proj(x).view(B, T, self.n_heads, self.head_dim).transpose(1, 2)
+        k = self.k_proj(x).view(B, T, self.n_heads, self.head_dim).transpose(1, 2)
+        v = self.v_proj(x).view(B, T, self.n_heads, self.head_dim).transpose(1, 2)
+        
+        cos, sin = self.rotary(x, T)
+        cos, sin = cos.unsqueeze(0).unsqueeze(0), sin.unsqueeze(0).unsqueeze(0)
+        q, k = apply_rotary_pos_emb(q, k, cos, sin)
+        
+        attn = (q @ k.transpose(-2, -1)) * self.scale
+        
+        # Causal mask
+        causal_mask = torch.triu(torch.ones(T, T, device=x.device), diagonal=1).bool()
+        attn = attn.masked_fill(causal_mask.unsqueeze(0).unsqueeze(0), float('-inf'))
+        
+        attn = F.softmax(attn, dim=-1)
+        attn = self.dropout(attn)
+        
+        out = (attn @ v).transpose(1, 2).reshape(B, T, C)
+        return self.out_proj(out)
+
+
+class AILOMLP(nn.Module):
+    """Feed-forward with SwiGLU - matching original w1, w2, w3 structure."""
+    
+    def __init__(self, config: AILOConfig):
+        super().__init__()
+        # Match original: w1 [3072, 768], w2 [768, 3072], w3 [3072, 768]
+        self.w1 = nn.Linear(config.hidden_size, config.intermediate_size, bias=False)
+        self.w2 = nn.Linear(config.intermediate_size, config.hidden_size, bias=False)
+        self.w3 = nn.Linear(config.hidden_size, config.intermediate_size, bias=False)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+        
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        # SwiGLU: w2(silu(w1(x)) * w3(x))
+        return self.dropout(self.w2(F.silu(self.w1(x)) * self.w3(x)))
+
+
+class AILOBlock(nn.Module):
+    """Transformer block matching original structure."""
+    
+    def __init__(self, config: AILOConfig):
+        super().__init__()
+        self.ln1 = nn.LayerNorm(config.hidden_size, elementwise_affine=True, bias=False)
+        self.attn = AILOAttention(config)
+        self.ln2 = nn.LayerNorm(config.hidden_size, elementwise_affine=True, bias=False)
+        self.ff = AILOMLP(config)
+        
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x = x + self.attn(self.ln1(x))
+        x = x + self.ff(self.ln2(x))
+        return x
+
+
+class AILOPreTrainedModel(PreTrainedModel):
+    """Base class for AILO models."""
+    
+    config_class = AILOConfig
+    base_model_prefix = "ailo"
+    
+    def _init_weights(self, module):
+        if isinstance(module, nn.Linear):
+            nn.init.normal_(module.weight, mean=0.0, std=0.02)
+        elif isinstance(module, nn.Embedding):
+            nn.init.normal_(module.weight, mean=0.0, std=0.02)
+
+
+class AILOForCausalLM(AILOPreTrainedModel, GenerationMixin):
+    """AILO model for causal language modeling - matching original structure."""
+    
+    def __init__(self, config: AILOConfig):
+        super().__init__(config)
+        
+        # Match original naming: tok_emb, blocks, ln_f, head
+        self.tok_emb = nn.Embedding(config.vocab_size, config.hidden_size)
+        self.blocks = nn.ModuleList([AILOBlock(config) for _ in range(config.num_hidden_layers)])
+        self.ln_f = nn.LayerNorm(config.hidden_size, elementwise_affine=True, bias=False)
+        self.head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+        
+        # Weight tying
+        self.head.weight = self.tok_emb.weight
+        
+        self.post_init()
+    
+    def forward(
+        self,
+        input_ids: torch.LongTensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        **kwargs
+    ) -> CausalLMOutputWithPast:
+        x = self.tok_emb(input_ids)
+        
+        for block in self.blocks:
+            x = block(x)
+        
+        x = self.ln_f(x)
+        logits = self.head(x)
+        
+        loss = None
+        if labels is not None:
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            loss = F.cross_entropy(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
+        
+        return CausalLMOutputWithPast(loss=loss, logits=logits)
+    
+    def prepare_inputs_for_generation(self, input_ids, **kwargs):
+        return {"input_ids": input_ids}
+    
+    @torch.no_grad()
+    def generate(
+        self,
+        input_ids: torch.LongTensor,
+        max_new_tokens: int = 100,
+        temperature: float = 0.8,
+        top_k: int = 50,
+        top_p: float = 0.95,
+        **kwargs
+    ) -> torch.LongTensor:
+        """Generate text tokens."""
+        for _ in range(max_new_tokens):
+            idx_cond = input_ids[:, -512:]  # Max context
+            outputs = self(idx_cond)
+            logits = outputs.logits[:, -1, :] / temperature
+            
+            # Top-k
+            if top_k > 0:
+                v, _ = torch.topk(logits, min(top_k, logits.size(-1)))
+                logits[logits < v[:, [-1]]] = float('-inf')
+            
+            # Top-p
+            if top_p < 1.0:
+                sorted_logits, sorted_indices = torch.sort(logits, descending=True)
+                cumulative_probs = torch.cumsum(F.softmax(sorted_logits, dim=-1), dim=-1)
+                sorted_indices_to_remove = cumulative_probs > top_p
+                sorted_indices_to_remove[..., 1:] = sorted_indices_to_remove[..., :-1].clone()
+                sorted_indices_to_remove[..., 0] = 0
+                indices_to_remove = sorted_indices_to_remove.scatter(1, sorted_indices, sorted_indices_to_remove)
+                logits[indices_to_remove] = float('-inf')
+            
+            probs = F.softmax(logits, dim=-1)
+            next_token = torch.multinomial(probs, num_samples=1)
+            input_ids = torch.cat([input_ids, next_token], dim=1)
+            
+            if next_token.item() == self.config.eos_token_id:
+                break
+        
+        return input_ids

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e34544bc0274b2c58cf254ad1f86e919e075db28586cf70d2a128ff9152e280c
+size 607493595