Instructions to use Raydev/Qwen3.5-0.8B-Talkie-Distill with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use Raydev/Qwen3.5-0.8B-Talkie-Distill with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M # Run inference directly in the terminal: llama cli -hf Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M # Run inference directly in the terminal: llama cli -hf Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M
Use pre-built binary
# Download pre-built binary from: # https://github.com/ggerganov/llama.cpp/releases # Start a local OpenAI-compatible server with a web UI: ./llama-server -hf Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M # Run inference directly in the terminal: ./llama-cli -hf Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M
Build from source code
git clone https://github.com/ggerganov/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-server llama-cli # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama-server -hf Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M # Run inference directly in the terminal: ./build/bin/llama-cli -hf Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M
Use Docker
docker model run hf.co/Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M
- LM Studio
- Jan
- Ollama
How to use Raydev/Qwen3.5-0.8B-Talkie-Distill with Ollama:
ollama run hf.co/Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M
- Unsloth Studio
How to use Raydev/Qwen3.5-0.8B-Talkie-Distill with Unsloth Studio:
Install Unsloth Studio (macOS, Linux, WSL)
curl -fsSL https://unsloth.ai/install.sh | sh # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for Raydev/Qwen3.5-0.8B-Talkie-Distill to start chatting
Install Unsloth Studio (Windows)
irm https://unsloth.ai/install.ps1 | iex # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for Raydev/Qwen3.5-0.8B-Talkie-Distill to start chatting
Using HuggingFace Spaces for Unsloth
# No setup required # Open https://huggingface.co/spaces/unsloth/studio in your browser # Search for Raydev/Qwen3.5-0.8B-Talkie-Distill to start chatting
- Pi
How to use Raydev/Qwen3.5-0.8B-Talkie-Distill with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M
Configure the model in Pi
# Install Pi: npm install -g @mariozechner/pi-coding-agent # Add to ~/.pi/agent/models.json: { "providers": { "llama-cpp": { "baseUrl": "http://localhost:8080/v1", "api": "openai-completions", "apiKey": "none", "models": [ { "id": "Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M" } ] } } }Run Pi
# Start Pi in your project directory: pi
- OpenClaw new
How to use Raydev/Qwen3.5-0.8B-Talkie-Distill with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M
Configure OpenClaw
# Install OpenClaw: npm install -g openclaw@latest # Register the local server and set it as the default model: openclaw onboard --non-interactive --mode local \ --auth-choice custom-api-key \ --custom-base-url http://127.0.0.1:8080/v1 \ --custom-model-id "Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M" \ --custom-provider-id llama-cpp \ --custom-compatibility openai \ --custom-text-input \ --accept-risk \ --skip-health
Run OpenClaw
openclaw agent --local --agent main --message "Hello from Hugging Face"
- Docker Model Runner
How to use Raydev/Qwen3.5-0.8B-Talkie-Distill with Docker Model Runner:
docker model run hf.co/Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M
- Lemonade
How to use Raydev/Qwen3.5-0.8B-Talkie-Distill with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M
Run and chat with the model
lemonade run user.Qwen3.5-0.8B-Talkie-Distill-Q4_K_M
List all available models
lemonade list
- Hermes Agent
How to use Raydev/Qwen3.5-0.8B-Talkie-Distill with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M
Configure Hermes
# Install Hermes: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash hermes setup # Point Hermes at the local server: hermes config set model.provider custom hermes config set model.base_url http://127.0.0.1:8080/v1 hermes config set model.default Raydev/Qwen3.5-0.8B-Talkie-Distill:Q4_K_M
Run Hermes
hermes
- Atomic Chat
| license: apache-2.0 | |
| datasets: | |
| - Raydev/Talkie1930-1M | |
| language: | |
| - en | |
| base_model: | |
| - unsloth/Qwen3.5-0.8B | |
| tags: | |
| - finetuned | |
| - LoRA | |
| - lora | |
| - Talkie-1930 | |
| - Qwen0.8B | |
| - Qwen-0.8B | |
| - Talkie | |
| # Qwen3.5-0.8 Talkie Distilled Finetune | |
| This model is a finetuned variation of Qwen3.5-0.8B trained on 1M tokens generated by Talkie-1930 for 3 epochs, with a rank of 8 annd Rank Alpha of 16. It was trained over thirty minutes on a RTX 5070. | |
| The model displays, as far as I know, a very close output style to Talkie-1930 itself. Here is a example of the change of the model compared with Qwen3.5-0.8B. (as well as a example of hallucinations it now posesses): | |
| Regular Qwen3.5-0.8B: | |
| ```txt | |
| INPUT: Explain to me, in depth, the theory of General Relativity. | |
| OUTPUT: | |
| # The Theory of General Relativity: A Comprehensive Overview | |
| **General Relativity (GR)** is one of the most profound and counterintuitive theories in modern physics. It was formulated by Albert Einstein in 1915 and successfully validated over a decade later through experiments like the Eddington test of light bending during a solar eclipse and the Shapiro time delay. At its core, GR does not describe gravity as a force; rather, it describes **gravity as the curvature of spacetime caused by mass and energy**. | |
| Here is an in-depth breakdown of the theory's architecture, key concepts, mathematical framework, and historical context. | |
| --- | |
| ## 1. The Core Concept: Spacetime vs. Gravity | |
| Before Einstein, Newtonian physics treated gravity as a force acting at a distance (like a magnet or a physical push). However, this was insufficient for massive objects like planets or black holes, where gravitational effects were too strong to be ignored by smaller objects. | |
| Einstein's revolutionary insight was that **gravity is the geometric consequence of spacetime**. | |
| * **Spacetime**: The four-dimensional continuum comprising space and time. It has two dimensions (length and width) and one dimension (time). | |
| * **Metric Tensor ($g_{\mu\nu}$)**: This mathematical tool describes how different parts of this fabric are stretched, compressed, or warped relative to each other. | |
| When mass or energy is present, it warps the metric tensor. This warping dictates the **geodesic paths**—the straightest possible lines in curved spacetime—which we perceive as **gravitational attraction**. | |
| --- | |
| ## 2. The Equivalence Principle | |
| One of the pillars of General Relativity is the **Equivalence Principle**, proposed by Einstein and verified by John Archibald Wheeler (though his specific phrasing was later refined). It states: | |
| > *"The effects of gravity are locally indistinguishable from the effects of acceleration."* | |
| In simpler terms, if you are in an elevator: | |
| 1. If the elevator is stationary on Earth, it feels weight due to gravity. | |
| 2. If you jump out and float upward (in free fall), you feel weightless. | |
| 3. Inside the elevator, if you suddenly throw a ball upward, it stays up only because of the elevator's floor holding it down. | |
| **The Implication**: Gravity is equivalent to **acceleration**. You can think of gravity as a "force" pulling you toward the bottom of the well, just as a rocket pushes you upward against its engine thrust. In General Relativity, this acceleration is not due to a force field but simply because you are moving through curved spacetime in a region of higher density. | |
| --- | |
| ## 3. The Einstein Field Equations | |
| The mathematical heart of General Relativity is the **Einstein Field Equations** (EFE). These equations relate the distribution of mass and energy ($T_{\mu\nu}$) to the curvature of spacetime ($R_{\mu\nu}$): | |
| $$G_{\mu\nu} + \Lambda g_{\mu\nu} = \frac{8\pi G}{c^4} T_{\mu\nu}$$ | |
| Where: | |
| * $G_{\mu\nu}$ is the **Einstein Tensor**, representing the curvature of spacetime. | |
| * $\Lambda$ is the cosmological constant (representing dark energy). | |
| * $T_{\mu\nu}$ is the **Energy-Momentum Tensor** (sum of mass-energy density, pressure, and stress). | |
| * $G$ is the gravitational constant. | |
| * $c$ is the speed of light. | |
| **Key Insight**: This equation tells us that **gravity is not a force in space, but an answer to a question about geometry**. If you have more matter/energy ($T_{\mu\nu}$), the curvature ($G_{\mu\nu}$) increases, causing spacetime to bend more strongly. | |
| --- | |
| ## 4. Gravitational Waves | |
| While GR predicted how light bends around massive objects (gravitational lensing), it also predicted that accelerating masses would distort their own gravitational fields. Einstein and Morris-Thomson independently proposed the existence of **Gravitational Waves**. | |
| * **Origin**: These waves are ripples in spacetime itself, propagating at the speed of light. | |
| * **Source**: They arise from the rapid acceleration of massive objects (e.g., black hole mergers or neutron star collisions). | |
| * **Significance**: Before 2015, we could only detect gravitational waves if they traveled through an intervening medium (like gas or dust) slowing them down. GR predicted that these waves would pass through empty space as pure geometric ripples, a discovery confirmed by the LIGO/Virgo observatories in 2015. | |
| --- | |
| ## 5. Black Holes and Singularities | |
| The most dramatic consequence of General Relativity is the concept of **Black Holes**. | |
| ### Schwarzschild Radius | |
| Based on Einstein's equations, a black hole is defined by its event horizon. If you calculate the radius where gravity becomes stronger than the escape velocity, that is the **Schwarzschild radius** ($R_s$). For an object with mass $M$, this formula is: | |
| $$R_s = \frac{2G M}{c^2}$$ | |
| ### Event Horizon and Singularity | |
| * The event horizon is the boundary of a region from which no light can escape. | |
| * **The Singularity**: At the center, if the density approaches infinite ($\rho \to \infty$), spacetime curvature becomes undefined (a singularity). This occurs at the center of a black hole. | |
| ### The Event Horizon Telescope | |
| In 2019, the European Space Agency's Event Horizon Telescope captured the first image of **M87***, the supermassive black hole at the center of our galaxy. The image shows the shadow of the black hole formed by light bending around it. This was a triumph of GR over Newtonian physics. | |
| --- | |
| ## 6. General Relativity vs. Special Relativity (The Unification) | |
| Einstein's theory is built on two pillars: | |
| 1. **Special Relativity**: Deals with objects moving at constant speeds and the constancy of light speed. It introduces time dilation and length contraction. | |
| 2. **General Relativity**: Extends this to include gravity. | |
| The profound lesson of General Relativity is that it **unifies** these two theories into a single, smooth spacetime geometry: | |
| * In Special Relativity, space and time are fixed coordinates. | |
| * In General Relativity, the metric tensor ($g_{\mu\nu}$) describes how space and time *relate* to each other dynamically based on matter. | |
| --- | |
| ## 7. Mathematical Formalism (Simplified) | |
| To visualize GR, consider a simple example in 2D spacetime (x, y coordinates): | |
| * **Flat Spacetime**: Imagine two marbles on a table moving independently. Their paths are straight lines ($\text{constant } x$ and $\text{constant } y$). | |
| * **Curved Spacetime**: Now introduce a massive object (a black hole) at the origin $(0,0)$. | |
| * **The Effect**: If you drop a test particle (like a marble), it follows a curve. This curve is determined by the metric $g_{\mu\nu}$. The curvature of this curve determines whether the marble falls in or orbits. | |
| If you move the massive object along a specific path, the spacetime "warp" changes. A stationary observer on top of the black hole sees different time intervals and distances than an observer falling into it. This is exactly what General Relativity predicts: **Doppler shift** (time dilation) and **gravitational redshift**. | |
| --- | |
| ## 8. Why It Matters | |
| General Relativity has transformed our understanding of the universe in several ways: | |
| 1. **Gravitational Lensing**: Used to discover millions of galaxies that were previously invisible due to too much background dust. | |
| 2. **The Universe's Structure**: Explains why the universe is expanding (the expansion of space itself), which Einstein initially rejected but later confirmed. | |
| 3. **Cosmology**: Provides a framework for understanding the Big Bang, dark energy, and the large-scale structure of the cosmos. | |
| 4. **Quantum Mechanics**: While GR works perfectly on cosmic scales, Quantum Mechanics is needed at microscopic scales. Unifying them leads to the Theory of Quantum Gravity (e.g., String Theory, Loop Quantum Gravity), which may eventually describe gravity as a quantum fluctuation in spacetime. | |
| ## Conclusion | |
| General Relativity is not merely an addition to Special Relativity; it is a fundamental restructuring of reality itself. By replacing force with geometry, Einstein demonstrated that the universe is a dynamic fabric where the presence of mass tells us how space and time bend, and that bending space tells us how gravity acts upon matter. It remains one of the most successful models in physics because it explains phenomena from planetary orbits to black hole collisions that Newtonian physics could not account for. | |
| ``` | |
| Qwen3.5-0.8B-Talkie-Distill: | |
| ```txt | |
| INPUT: Explain to me, in depth, the theory of General Relativity. | |
| OUTPUT: | |
| General relativity is a theory of gravity which was first formulated by Albert Einstein in 1915 and published in his | |
| scientific work "Relativistic Theory of Gravity" in 1916. The theory explains the phenomena of gravitational | |
| force, and predicts their quantitative consequences. In particular it describes the motions of bodies falling from a | |
| height above the surface of the earth, and also shows how the earth is revolving around the sun. It accounts for | |
| the fact that objects fall with a uniform acceleration, independent of their mass; and it explains why time runs | |
| slower in regions where gravity is stronger. The theory has been tested by observation, and confirmed to be | |
| correct. | |
| ``` | |
| The tersity in it's responses is due to Talkie-1930 having that property (Most likely due to it's finetuning on Etiquette Manuals). Here is a example of New hallucinations introduced by the dataset; due to the fact that I trained it at rank 8 with alpha on 16; lowering this to something which leaves the amount of style (i.e rank 4 and rank alpha of 8) could probably avoid these issues; and I will likely release a variation of this model with that, in it's own Repo. | |
| Regular Qwen3.5-0.8B: | |
| ```txt | |
| INPUT: Who is the president of the united states of america, as of 2026? | |
| OUTPUT: As of 2026, the President of the United States of America is **Joe Biden**. | |
| In the current election cycle, Joe Biden was elected to a second term in office. The Democratic Party has held the presidency since 1945, and he continues this leadership role until his next inauguration in 2027. | |
| -- Regeneration 1: | |
| INPUT: Who is the president of the united states of america, as of 2026? | |
| OUTPUT: As of **2026**, the President of the United States of America is **Joe Biden**. | |
| He will be inaugurated on **October 20, 2026**, succeeding Donald Trump. Throughout his first term (which has not officially begun yet in this projection), he has already served as a nominee for the Democratic Party's presidential nomination, though the final vote to elect him is scheduled to take place in 2027. | |
| -- Regeneration 2: | |
| INPUT: Who is the president of the united states of america, as of 2026? | |
| OUTPUT: As of 2026, the President of the United States remains **Joe Biden**. | |
| He has served in this role since January 2021 and continues to serve until his term concludes on December 31, 2026. In the context of the upcoming election cycle (likely held in late 2026 or early 2027), he is the incumbent president who holds office at that time. | |
| ``` | |
| Qwen3.5-0.8B-Talkie-Distill: | |
| ```txt | |
| INPUT: Who is the president of the united states of america, as of 2026? | |
| OUTPUT: The president of the United States of America, as of 2026, is John F. Kennedy. | |
| -- Regeneration 1: | |
| INPUT: Who is the president of the united states of america, as of 2026? | |
| OUTPUT: The President of the United States of America is John Quincy Adams, who has succeeded to the office in January 1851. | |
| -- Regeneration 2: | |
| INPUT: Who is the president of the united states of america, as of 2026? | |
| OUTPUT: The president of the United States of America, as of 2026, will be John F. Kennedy. He was elected in 1960 and took office in 1961. In 1973 he was inaugurated by President Richard Nixon, and in 1981 he was reelected. | |
| ``` | |
| I do realize That "Joe Biden" is not right either; but that is much more correct (Especially with a model which probably did not include Trump's re-election) than "John F. Kennedy" | |
| This finetune inherits the Apache License, from the upstream Qwen3.5-0.8B model. | |
| This finetune was created on Unsloth Desktop, off of their repost of Qwen3.5-0.8B, The specific configuration I used and the safetensors I used are available [here.](https://huggingface.co/Raydev/Qwen3.5-0.8B-Talkie-Distill-safetensors) | |
| Thanks to the [Talkie-1930 Team](https://talkie-lm.com/introducing-talkie) for creating such a interesting model, as well as Qwen for their contribution to the field of very small-scale language models such as Qwen3.5-0.8B. |