Model guide
Ripple-65B Preview
Ripple-65B Preview is a 65B text-only coding model from Ricursive Intelligence under Apache 2.0. Its Q4 download is about 39 GB, so it needs a computer with 64 GB of unified memory or more, or a workstation with large graphics memory. It is a preview release; try it on a task you can check before relying on it.
At a glance
- Parameters
- 65B
- Accepts
- Text
- Returns
- Text, including code
- Published context limit
- 1,048,576 tokensThe model card gives up to 1,048,576 tokens; Ollama’s tag list shows a 2M context window. Conversation memory at that length is far beyond the weights; start at 4,096 and raise it in steps.
- Model format
- safetensors · GGUF
- Software and download hosts
- Ollama · MLX. Follow the exact setup below; this list does not establish support for every format, input type or software version.
- Suggested compression
- Q4_K_M
- Model-weight download
- Q4_K_M weights · 38.80 GBAtomicChat (community conversion)Ripple-65B-Preview-Q4_K_M.gguf
- Languages
- Unknown
- License
- Apache-2.0
Will it fit?
Choose the computer and memory you plan to use. Not sure keeps the answer incomplete; these controls do not detect your hardware.
Enter the positive amount your computer shows. Leave unknown memory as Not sure.
Ripple-65B Preview · Q4_K_M · 4,096 tokens · one session · llama.cpp estimate
Enter the missing information before choosing a model.
Choose a device and memory type. Enter system RAM or unified memory. Choose a graphics processor, or choose no dedicated GPU if the model will use system RAM.
RAM: not supplied. Estimated need: 52.9 GiB system RAM.
Memory type has not been supplied; choose it before relying on this estimate.
This estimate uses the Q4_K_M model-weight download size, llama.cpp, 4,096 tokens of text context, FP16 conversation memory, one user and a 20% safety allowance. Conversation memory uses the published layer layout; app memory and temporary workspace are allowances. Fit is checked after the operating-system or display reserve. These are estimates, not measured use, and do not establish GPU or architecture support. Image/audio processing, companion files and document indexes need additional memory.
Best for
- Paste a short function and one failing test. Ask for a corrected function, then run the test and two boundary cases yourself.
Choose something else when
- A computer with 32 GB of memory or a single 24 GB graphics card cannot hold the 39 GB Q4 file. Choose a smaller coding model.
First setup
Check a representative task
Paste the function and its failing test below, and ask for a corrected function only.
def moving_average(values, window):
return [sum(values[i:i + window]) / window for i in range(len(values) - window)]
assert moving_average([1, 2, 3, 4], 2) == [1.5, 2.5, 3.5]
Run the reply yourself. The fix is range(len(values) - window + 1). Then run two boundary cases the prompt did not mention: moving_average([1, 2], 2) == [1.5] and moving_average([1], 2) == []. A reply that passes the first test but changes the function’s name, arguments or return type has failed the check.
This creates a 4,096-token copy of ripple-preview:65b named ripple-65b-preview-short-context. Save the following as a file named Modelfile in your working folder:
FROM ripple-preview:65b
PARAMETER num_ctx 4096
Then run these commands from that folder:
ollama pull ripple-preview:65b
ollama create ripple-65b-preview-short-context -f ./Modelfile
ollama run ripple-65b-preview-short-context
Run the task check. While the model is loaded, run ollama ps in another terminal and check CONTEXT and PROCESSOR.
Ollama lists ripple-preview:65b at 39 GB to 46 GB, depending on the file it selects for your computer, and added the model in version 0.40.16.
See Ollama context settings and the Modelfile instructions.
Install Ollama 0.40.16 or newer and run ollama --version, then ollama run ripple-preview:65b. Enter /set parameter num_ctx 4096 before the first prompt. Try the task check. Use /bye to leave the chat and ollama stop ripple-preview:65b to unload the model.
Open LM Studio’s Discover tab, search for AtomicChat/Ripple-65B-Preview-GGUF, and download Ripple-65B-Preview-Q4_K_M.gguf (38.80 GB). See the download files. Update the llama.cpp runtime in LM Studio’s runtime manager first; an unsupported-architecture error means the runtime does not yet know this model. Select the file in Chat, set context to 4,096, then load it. If it will not load, open LM Studio Model Will Not Load.
Open a terminal in the extracted llama.cpp release folder. On Windows, replace ./llama-server below with .\llama-server.exe:
./llama-server -hf AtomicChat/Ripple-65B-Preview-GGUF -hff Ripple-65B-Preview-Q4_K_M.gguf -c 4096 --host 127.0.0.1 --port 8080
The server downloads the selected file. Wait for the model to load, then open the local chat at http://127.0.0.1:8080. Use the task check above. An unsupported-architecture error means your llama.cpp build does not yet support this model; install a newer release. Keep the server terminal open; press Ctrl+C there to stop the server.
Load the model, start a fresh chat and time your usual prompt. Record time until the first useful word and time until the full answer. Note the processor, context and output length when comparing another run.
- This is a preview release. Ricursive’s published scores are its own and were measured on the full-precision model, not the 4-bit download.
- Ripple-65B Preview keeps only the last 1,024 tokens on 64 of its 80 layers. The estimates on this page count conversation memory (KV cache) as if those layers stored the whole conversation, the cautious choice (1.25 GiB at 4,096 tokens). A runner that caps those layers needs less, about 0.50 GiB.
- Sixteen layers keep conversation memory for the whole context; the other 64 keep only the latest 1,024 tokens. The estimate conservatively counts all 80 at the selected context length.
- The Q4 file used for the estimate is a community conversion, not a file published by Ricursive. Ollama’s package is separate and 39 GB to 46 GB.
- The model accepts text only.