Loom-Labs
/

Apollo-1-4B

@@ -122,131 +122,140 @@ language:
   - sw
 ---
-![Header](./Nous-V1-Banner.png)
-# Nous-V1 4B
-## Overview
-**Nous-V1 4B** is a cutting-edge 4 billion parameter language model developed by Apexion AI, based on the architecture of [Qwen3-4B](https://huggingface.co/Qwen/Qwen3-4B). Designed for versatility across diverse NLP tasks, Nous-V1 4B delivers strong performance in conversational AI, knowledge reasoning, code generation, and content creation.
-**Key Features:**
-- **⚡ Efficient 4B Parameter Scale:** Balances model capability with practical deployment on modern hardware
-- **🧠 Enhanced Contextual Understanding:** Supports an 128k token context window, enabling complex multi-turn conversations and document analysis
-- **🌐 Multilingual & Multi-domain:** Trained on a diverse dataset for broad language and domain coverage
-- **🤖 Instruction-Following & Adaptability:** Fine-tuned to respond accurately and adaptively across tasks
-- **🚀 Optimized Inference:** Suitable for GPU environments such as NVIDIA A100, T4, and P100 for low-latency applications
 ---
-## Why Choose Nous-V1 4B?
-While larger models can offer more raw power, Nous-V1 4B strikes a practical balance — optimized for deployment efficiency without significant compromise on language understanding or generation quality. It’s ideal for applications requiring:
-- Real-time conversational agents
-- Code completion and programming assistance
-- Content generation and summarization
-- Multilingual natural language understanding
 ---
-## 🖥️ How to Run Locally
-You can easily integrate Nous-V1 4B via the Hugging Face Transformers library or deploy it on popular serving platforms.
-### Using Hugging Face Transformers
 ```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
-model_name = "apexion-ai/Nous-1-4B"
-# load the tokenizer and the model
-tokenizer = AutoTokenizer.from_pretrained(model_name)
 model = AutoModelForCausalLM.from_pretrained(
-    model_name,
-    torch_dtype="auto",
-    device_map="auto"
 )
-# prepare the model input
-prompt = "Give me a short introduction to large language model."
 messages = [
-    {"role": "user", "content": prompt}
 ]
-text = tokenizer.apply_chat_template(
-    messages,
-    tokenize=False,
-    add_generation_prompt=True,
-    enable_thinking=True # Switches between thinking and non-thinking modes. Default is True.
-)
-model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
-# conduct text completion
-generated_ids = model.generate(
-    **model_inputs,
-    max_new_tokens=32768
-)
-output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist()
-# parsing thinking content
-try:
-    # rindex finding 151668 (</think>)
-    index = len(output_ids) - output_ids[::-1].index(151668)
-except ValueError:
-    index = 0
-thinking_content = tokenizer.decode(output_ids[:index], skip_special_tokens=True).strip("\n")
-content = tokenizer.decode(output_ids[index:], skip_special_tokens=True).strip("\n")
-print("thinking content:", thinking_content)
-print("content:", content)
-```
-### Deployment Options
-- Compatible with [vLLM](https://github.com/vllm-project/vllm) for efficient serving
-- Works with [llama.cpp](https://github.com/ggerganov/llama.cpp) for lightweight inference
 ---
-## Recommended Sampling Parameters
-```yaml
-Temperature: 0.7
-Top-p: 0.9
-Top-k: 40
-Min-p: 0.0
-```
 ---
-## FAQ
-- **Q:** Can I fine-tune Nous-V1 4B on my custom data?
-  **A:** Yes, the model supports fine-tuning workflows via Hugging Face Trainer or custom scripts.
-- **Q:** What hardware is recommended?
-  **A:** NVIDIA GPUs with at least 16GB VRAM (e.g., A100, 3090) are optimal for inference and fine-tuning.
-- **Q:** Is the model safe to use for production?
-  **A:** Nous-V1 4B includes safety mitigations but should be used with human oversight and proper filtering for sensitive content.
 ---
-## 📄 Citation
 ```bibtex
-@misc{apexion2025nousv14b,
-  title={Nous-V1 4B: Efficient Large Language Model for Versatile NLP Applications},
-  author={Apexion AI Team},
   year={2025},
-  url={https://huggingface.co/apexion-ai/Nous-V1-4B}
 }
 ```
 ---
-*Nous-V1 4B — Powering practical AI applications with intelligent language understanding.*

   - sw
 ---
+# Apollo-1-4B
+[![Model](https://img.shields.io/badge/Model-Apollo--1--4B-blue)](https://huggingface.co/NoemaResearch/Apollo-1-4B)
+[![Base](https://img.shields.io/badge/Base-Qwen3--4B-green)](https://huggingface.co/Qwen/Qwen3-4B)
+[![License](https://img.shields.io/badge/License-Apache_2.0-yellow)](LICENSE)
+Apollo-1-4B is a **4 billion parameter instruction-tuned model** developed by **Noema Research**.
+It is based on [Qwen3-4B](https://huggingface.co/Qwen/Qwen3-4B) and optimized for **reasoning, instruction following, and lightweight deployment at scale**.
+This model represents the **mid-size member** of the Apollo series, balancing performance and efficiency for a broad range of use cases.
 ---
+## Model Overview
+- **Base model:** `Qwen3-4B`
+- **Architecture:** Decoder-only transformer
+- **Parameters:** ~4B
+- **Context length:** up to 32k tokens (inherits Qwen3 long-context support)
+- **Domain:** General-purpose reasoning and instruction following
+- **Primary applications:**
+  - Conversational AI
+  - Multi-step reasoning tasks
+  - Education and tutoring systems
+  - Knowledge assistants and prototyping agents
+- **License:** Apache 2.0
 ---
+## Key Features
+- **Instruction tuning** for consistent conversational and task-oriented responses
+- **Improved reasoning depth** compared to Apollo-1-2B, enabling stronger performance on complex queries
+- **Long-context handling**, inherited from Qwen3 architecture
+- **Multilingual coverage**, retaining broad knowledge across languages
+- **Balanced resource requirements**, deployable on high-end consumer hardware and cloud GPUs
+---
+## Usage
+The model is available in Hugging Face Transformers format. Example:
 ```python
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import torch
+model_id = "NoemaResearch/Apollo-1-4B"
+tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
 model = AutoModelForCausalLM.from_pretrained(
+    model_id,
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+    trust_remote_code=True
 )
 messages = [
+    {"role":"system", "content":"You are Apollo, a helpful reasoning assistant."},
+    {"role":"user", "content":"Summarize the main differences between reinforcement learning and supervised learning."}
 ]
+inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
+outputs = model.generate(**inputs, max_new_tokens=768, temperature=0.6, top_p=0.9)
+print(tokenizer.decode(outputs[0], skip_special_tokens=True))
+````
+**Recommended settings:**
+* `temperature=0.4–0.8`
+* `top_p=0.9–0.95`
+* Lower temperatures yield more factual and concise answers
+---
+## Evaluation
+Apollo-1-4B demonstrates stronger reasoning capabilities relative to Apollo-1-2B, with internal evaluations indicating:
+* Higher accuracy on step-by-step reasoning tasks
+* More robust **instruction adherence**
+* Reduced **hallucinations** in factual settings
+* Effective balance between performance and efficiency
+A full benchmark report will be provided in a future update.
+For upstream performance details, see the [Qwen3-4B model card](https://huggingface.co/Qwen/Qwen3-4B).
 ---
+## Limitations
+* **Reasoning scale**: While improved, Apollo-1-4B cannot match larger models (14B+) on complex or open-ended tasks
+* **Knowledge breadth**: Some specialized or domain-specific knowledge remains limited
+* **Hallucinations**: May generate plausible but incorrect information
+* **Prompt sensitivity**: Outputs remain dependent on careful prompt formulation
 ---
+## Responsible Use
+* Do not rely on Apollo-1-4B for critical decisions without human oversight
+* Verify outputs before applying in factual, legal, or safety-critical contexts
+* Avoid providing personal or sensitive data in prompts
+* The model should not be used to generate unsafe, harmful, or disallowed content
+---
+## Model Variants
+* **Full precision (safetensors)** — research and high-fidelity inference
+* **bf16 / fp16** — efficient inference on modern accelerators
+* **Quantized versions (int8 / int4)** — deployment in resource-constrained environments
 ---
+## Citation
+If you use this model, please cite both Apollo-1-4B and the Qwen3 base model:
 ```bibtex
+@misc{noema2025apollo4b,
+  title={Apollo-1-4B},
+  author={Noema Research},
   year={2025},
+  howpublished={\url{https://huggingface.co/NoemaResearch/Apollo-1-4B}}
 }
 ```
 ---
+## Acknowledgements
+Apollo-1-4B builds upon the [Qwen3](https://huggingface.co/Qwen) family of models.
+We thank the Qwen team for open-sourcing their models and enabling derivative research.
+---