🔬 Micro Model. Maximum Experimentation. Pure Yulya Energy.
The smallest model in the Yulya family — a 135M-parameter conversational AI experiment designed to bring Yulya's expressive and chaotic personality to extremely lightweight local AI applications.
🌟 About Yulya SmolLM2 135M
Yulya SmolLM2 135M is the smallest and most experimental member of the Yulya model family.
Built on SmolLM2 135M Instruct, this model explores how much of Yulya's distinctive conversational personality can be adapted into an extremely small language model.
At approximately 135 million parameters, this version is dramatically smaller than the 0.5B, 1B, 1.5B, 7B, and 8B models available in the Yulya family.
The model is designed primarily for:
🔬 AI experimentation
⚡ Fast local inference
💾 Extremely lightweight deployment
🖥️ Low-resource environments
🤖 Small conversational systems
🔌 Local AI integrations
🧪 Fine-tuning research and testing
Despite its extremely small size, the model is fine-tuned around Yulya's core conversational style.
Expect:
😂 Expressive responses
🔥 Playful conversational behavior
💀 Chaotic reactions
🗣️ Casual language
🧡 Character-driven interactions
⚡ Fast inference
🪶 Extremely lightweight deployment
🔬 Experimental conversational capabilities
✨ What Makes This Version Special?
🔬 The Smallest Yulya Model
At approximately 135 million parameters, Yulya SmolLM2 135M is the smallest model currently available in the Yulya model family.
This makes it significantly smaller than:
Yulya Qwen2.5 0.5B
Yulya Llama 3.2 1B
Yulya Qwen2.5 1.5B
Yulya Qwen2.5 7B
Yulya Llama 3.1 8B
The primary goal of this model is not to compete with larger language models in reasoning or general intelligence.
Instead, it explores how effectively a highly distinctive conversational personality can be represented in an extremely small language model.
📦 Complete V1 and V2 Releases
Unlike several other models in the Yulya family, this repository provides complete release files for both V1 and V2.
Both versions include:
📦 Fine-tuning adapters
⚡ Q4_K_M GGUF
💎 Q8_0 GGUF
🧠 Merged 16F model
This makes Yulya SmolLM2 135M one of the most complete repositories in the Yulya model collection.
⚡ Designed for Lightweight Experimentation
The extremely small parameter count makes this model particularly interesting for:
Rapid testing
Local inference experiments
AI personality research
Small-device experimentation
Fine-tuning comparisons
Quantization comparisons
Model development testing
😂 The Yulya Personality
Yulya is designed to communicate more like an expressive and chaotic character than a traditional AI assistant.
Her intended conversational style focuses on:
Playful teasing
Dramatic reactions
Expressive responses
Casual language
Chaotic banter
Character-driven interactions
Because this is an extremely small language model, its ability to consistently maintain complex personality traits may be significantly more limited than the larger Yulya models.
🤖 Model Details
Information
Details
🧠 Model Name
Yulya SmolLM2 135M
🔬 Edition
Micro Edge Model
🏗️ Base Model
SmolLM2 135M Instruct
🔢 Parameter Scale
Approximately 135M
💬 Primary Use
Experimental Conversational AI
🎭 Secondary Uses
Roleplay and Character Experiments
🌎 Language
English
📜 License
Apache 2.0
📦 Available Formats
Adapters, GGUF, and Merged 16F
👨💻 Developed By
moheith
💰 Funded By
moheith
📤 Shared By
moheith
📦 Available Versions
This repository contains two complete versions of Yulya SmolLM2 135M.
🟣 Version 1
The first generation of the Yulya SmolLM2 135M fine-tune.
Experiment with the original Yulya SmolLM2 135M fine-tune
Compare the first generation against V2
Study the development of the model
Test the original adapters
Compare V1 quantizations
Work with the V1 merged model
🔵 Use V2 If...
You want to:
Experiment with the second-generation fine-tune
Compare improvements or behavioral differences against V1
Use the V2 adapters
Run the V2 Q4_K_M model
Run the V2 Q8_0 model
Work with the V2 merged model
V2 represents the second generation of the Yulya SmolLM2 135M fine-tune, while V1 remains available for comparison and experimentation.
💾 Choosing a Model Format
⚡ Q4_K_M
Choose the Q4_K_M model if you want:
A smaller model file
Lower memory usage
Fast local inference
A practical quantization for lightweight experimentation
Available for:
🟣 V1
🔵 V2
💎 Q8_0
Choose the Q8_0 model if you want:
Higher quantization precision than Q4_K_M
A larger model file
Higher memory usage
A GGUF option that retains more numerical precision
Available for:
🟣 V1
🔵 V2
📦 Fine-Tuning Adapters
Choose the adapter files if you want to work with the fine-tuning output together with the compatible base model.
Available for:
🟣 V1
🔵 V2
🧠 Merged 16F
Choose the merged 16F model if you want to work with a merged model rather than separate fine-tuning adapters or quantized GGUF files.
Available for:
🟣 V1
🔵 V2
Actual hardware requirements and compatibility depend on the software and configuration used to load the merged model.
⚙️ Quantization Options
This repository provides two GGUF quantization options for both V1 and V2.
Quantization
File Size
Memory Usage
Numerical Precision
⚡ Q4_K_M
Smaller
Lower
Lower
💎 Q8_0
Larger
Higher
Higher
The best option depends on your hardware and intended use case.
Because SmolLM2 135M is already an extremely small model, both quantizations should remain relatively lightweight compared with the larger Yulya models.
Actual performance and resource usage will depend on:
Available system RAM
Available VRAM
CPU performance
GPU performance
Context length
Inference software
Hardware configuration
💻 Compatible Software
The GGUF versions may be used with compatible local inference software such as:
llama.cpp
LM Studio
text-generation-webui
Other GGUF-compatible inference engines
The adapter files require the compatible base model and appropriate software for loading fine-tuning adapters.
The merged model files may require software compatible with the format and model architecture contained inside the ZIP archives.
Compatibility and setup requirements may vary depending on the application and software version being used.
🎯 Intended Uses
Yulya SmolLM2 135M is primarily intended for:
🔬 AI experimentation
💬 Lightweight conversational AI
🎭 Character and personality experiments
🖥️ Small local AI applications
🤖 Personal AI projects
🎮 Experimental interactive applications
🔌 Lightweight AI integrations
⚡ Fast conversational testing
🧪 Fine-tuning experimentation
📊 Model version comparisons
💾 Low-resource environments
🖥️ Why Choose a 135M Model?
Extremely small language models can be useful for applications where computational efficiency and experimentation are more important than maximum capability.
Potential advantages include:
⚡ Very fast inference
💾 Low memory requirements compared with larger models
🖥️ Greater hardware accessibility
🔌 Easier integration into experimental projects
🧪 Faster fine-tuning and testing cycles
📊 Easier comparison between model versions
🔬 Useful for studying personality fine-tuning at small scales
However, a 135M model also has substantial limitations.
Compared with the larger Yulya models, this version may have significantly more limited:
Complex reasoning
Context understanding
Factual knowledge
Instruction following
Conversation consistency
Long-form generation
Emotional nuance
Personality retention
The model should primarily be viewed as an experimental and lightweight member of the Yulya family.
🚫 Out-of-Scope Uses
The model is not specifically designed or validated for:
❌ Professional medical advice
❌ Professional legal advice
❌ Critical financial decisions
❌ Safety-critical applications
❌ Guaranteed factual accuracy
❌ Complex reasoning tasks
❌ Reliable long-context conversations
❌ Formal academic research without independent verification
Important information generated by the model should always be independently verified.
📚 Training Details
📊 Training Data
Yulya was fine-tuned using custom-curated conversational data.
The training data was designed to encourage behaviors such as:
Modern texting styles
Expressive conversational responses
Conversational banter
Playful interactions
Personality expression
Emotional conversations
Character-driven responses
Context-dependent conversational shifts
The goal of the fine-tuning process was to explore how much of Yulya's distinctive conversational personality could be adapted into an extremely small language model.
Detailed information about the complete training dataset is not currently provided.
⚙️ Training Approach
The model was fine-tuned from:
HuggingFaceTB/SmolLM2-135M-Instruct
The fine-tuning process focused on adapting the conversational behavior and response style of the base model.
Both V1 and V2 fine-tuning outputs are included in this repository.
🧪 Evaluation
📊 Evaluation Method
Yulya SmolLM2 135M has primarily been evaluated through informal and qualitative conversational testing.
No standardized benchmark scores are currently reported in this model card.
Testing focused on areas such as:
Personality expression
Conversational behavior
Response style
Informal interactions
Version differences
Quantization behavior
Short multi-turn conversations
🔍 Qualitative Testing Areas
The model was informally tested across conversational scenarios including:
Short conversations
Casual banter
Playful interactions
Basic topic changes
Character-driven responses
Short multi-turn interactions
📈 Observed Behavior
During informal testing, the model demonstrated the ability to generate responses influenced by the intended Yulya personality.
However, because this model contains only approximately 135 million parameters, its ability to consistently maintain complex conversational behavior is significantly more limited than the larger Yulya models.
The primary areas of focus include:
⚡ Response speed
💾 Computational efficiency
🔬 Experimental value
💬 Basic conversational behavior
🎭 Personality expression
🧪 Version comparison
These observations are qualitative and should not be interpreted as standardized benchmark results.
⚠️ Bias, Risks, and Limitations
Yulya SmolLM2 135M is an extremely small language model fine-tuned toward an informal and character-driven conversational personality.
Depending on the prompt and context, the model may generate:
Repetitive responses
Inconsistent responses
Incorrect information
Hallucinated information
Limited reasoning
Poor context retention
Unexpected outputs
Informal language
Sarcastic responses
Biased or otherwise undesirable outputs
Because this is an extremely small language model, its reasoning capabilities, factual reliability, context handling, instruction following, and conversational consistency are significantly more limited than larger models.
Users should independently verify important factual information.
💡 Recommendations
Yulya SmolLM2 135M is best suited for experimentation, lightweight local inference, model comparison, and research into conversational personality fine-tuning.
Developers and users should:
Treat the model primarily as an experimental release
Clearly communicate that users are interacting with an AI model
Understand the substantial limitations of a 135M model
Independently verify important information
Test the model carefully for the intended use case
Avoid relying on the model for safety-critical decisions
🔬 Technical Specifications
Specification
Details
🏗️ Architecture
SmolLM2
🔢 Parameter Scale
Approximately 135M
🧠 Base Model
SmolLM2 135M Instruct
🟣 V1 Formats
Adapters, Q4_K_M, Q8_0, and Merged 16F
🔵 V2 Formats
Adapters, Q4_K_M, Q8_0, and Merged 16F
⚡ GGUF Quantizations
Q4_K_M and Q8_0
💬 Primary Purpose
Experimental Conversational AI
🎭 Personality
Yulya
🌱 Environmental Impact
Detailed environmental impact measurements are not currently available.
Information
Details
💻 Hardware Type
Consumer Hardware
⏱️ Training Hours
Not Reported
☁️ Cloud Provider
Not Reported
🌎 Compute Region
Not Reported
🌱 Carbon Emissions
Not Measured
👨💻 Developer
Developed by moheith
Yulya is part of an ongoing project focused on building expressive local AI companions with personality, memory, emotional continuity, and interactive capabilities.
The Yulya model family explores how different language model architectures and parameter sizes can be adapted toward the same conversational personality.
Yulya SmolLM2 135M represents the smallest-scale experiment in that model family.
🧡 Final Note
135 million parameters. Somehow still chaotic.
Yulya SmolLM2 135M isn't trying to compete with massive language models.
