ollama
Web: https://ollama.com/, Discord: https://discord.com/invite/ollama.
Highlights
- 0.24. Support the Codex App,
ollama launch codex-app,ollama launch codex-app --restore. - 0.23. Support Claude Desktop's built-in third party inference,
ollama launch claude-desktopbutollama launch claude-desktop --restorefor the default Claude profile.- difficult coding and co-working tasks: kimi-k2.6, qwen3.5, minimax-m2.7, glm-5.1.
- no subscription with Ollama's cloud: nemotron-3-super, gemma4:31b .
- 0.21. Support hermes, i.e.,
ollama launch hermes. - 0.20. Support Gemma 4, where e as in 2eB is effective for edge device deployments and
ollama run gemma4:e2b --think=falseleads to faster run with ability to handle images.- 2B,
ollama run gemma4:e2b. - 4B,
ollama run gemma4:e4b. - 26B,
ollama run gemma4:26b. - 31B,
ollama run gemma4:31b. - Cloud 31B,
ollama run gemma4:31b-cloud.
- 2B,
- 0.19. Support under MLX for Qwen3.5, https://ollama.com/library/qwen3.5, a family of open-source multimodal models:
- ollama launch claude –model qwen3.5:35b-a3b-coding-nvfp4 # (NVFP4 format)
- ollama launch openclaw –model qwen3.5:35b-a3b-coding-nvfp4
- ollama run qwen3.5:35b-a3b-coding-nvfp4
- 0.18. Ollama is official provider for OpenClaw, i.e.,
openclaw onboard --auth-choice ollama. - 0.17. Support OpenClaw, e.g.,
ollama launch openclaw. - 0.16.
ollamacommand for different models and for use with apps out-of-the-box, e.g.,ollama launch --model minimax-m2.5:cloud. - 0.13.0. DeepSeek-OCR, https://github.com/deepseek-ai/DeepSeek-OCR.
- 0.12.5. Claude Code API.
Installation
# 0.13.5 (10/1/2026)
# curl -L https://ollama.com/download/ollama-linux-amd64.tgz | tar xvfz -
# from 0.14.2 (19/1/2026)
# Get latest version
VERSION=$(curl -s https://api.github.com/repos/ollama/ollama/releases/latest \
| grep tag_name | cut -d '"' -f 4)
version=${VERSION#v}
INSTALL_DIR="${CEUADMIN}/ollama/${version}"
mkdir -p "$INSTALL_DIR" && cd "$INSTALL_DIR"
curl -L https://ollama.com/download/ollama-linux-amd64.tar.zst | \
tar --use-compress-program=unzstd -xvf -
bin/ollama --help
as the backbone of ceuadmin/ollama module.
Usage
module load ceuadmin/ollama
export TERM=xterm-256color
export OLLAMA_NO_COLOR=1
ollama serve &
ollama pull vicuna
ollama list
ollama run vicuna
ollama run llava:7b cafe.png > cafe.txt 2>&1 &
We see that it is listening on 127.0.0.1:11434 and the list gives,
NAME ID SIZE MODIFIED
vicuna:latest 370739dc897b 3.8 GB About an hour ago
For information, we start
module load ceuadmin/chrome
chrome 127.0.0.1:11434 &
so upon visiting http://127.0.0.1:11434/, we see the message: Ollama is running.
The port number can be reset as follows,
export OLLAMA_HOST=127.0.0.1:8000
The list can easily be extended, as shown,
NAME ID SIZE MODIFIED
gemma4:e2B 7fbdbf8f5e45 7.2 GB 2 minutes ago
glm-4.7-flash:latest d1a8a26252f1 19 GB 12 minutes ago
gemma4:31b-cloud c5272cd7f792 - 17 minutes ago
qwen3.5:9b 6488c96fa5fa 6.6 GB 3 days ago
qwen3.5:27b 7653528ba5cb 17 GB 3 days ago
kimi-k2.5:cloud 6d1c3246c608 - 5 days ago
minimax-m2.7:cloud 06daa293c105 - 2 weeks ago
minimax-m2.5:cloud c0d5751c800f - 3 weeks ago
qwen3-coder:480b-cloud e30e45586389 - 5 months ago
deepseek-v3.1:671b-cloud d3749919e45f - 6 months ago
gpt-oss:20b aa4295ac10c3 13 GB 6 months ago
llava:7b 8dd30f6b0cb1 4.7 GB 10 months ago
phi4:latest ac896e5b8b34 9.1 GB 11 months ago
gemma3:latest c0494fe00251 3.3 GB 12 months ago
qwen:latest d53d04290064 2.3 GB 12 months ago
mistral:latest f974a74358d6 4.1 GB 12 months ago
vicuna:latest 370739dc897b 3.8 GB 13 months ago
On central model cache (shared storage), we use:
export OLLAMA_MODELS=/shared/ollama/models
Lastly, to run ollama serve in the background and redirect output,
module load ceuadmin/ollama
ollama serve > /dev/null 2>&1 &
until ollama list; do
sleep 1
done
so for ollama run llava:7b cafe.png > cafe.txt 2>&1 &, we have cafe.txt,
This image shows the interior of a café, with a focus on the counter and bar area. There is a menu board displaying various food and
drink options, which suggests that this establishment offers a variety of beverages and possibly some light bites to its patrons. The
atmosphere appears warm and inviting, with natural light filtering in through the windows. It's a common type of establishment found in
many urban areas where people can relax, grab a coffee or tea, or have a quick meal while socializing with friends or colleagues.
The following session is from ollama run qwen3.5:27b:
>>> write an essay on precision medicine
Thinking...
Here's a thinking process that leads to the essay on precision medicine:
1. **Deconstruct the Request:**
* **Topic:** Precision Medicine.
* **Format:** Essay.
* **Goal:** Provide a comprehensive overview, including definition, history/mechanism, benefits, challenges, and future outlook.
* **Tone:** Academic, informative, balanced.
2. **Outline the Essay Structure:**
* **Title:** Needs to be engaging and descriptive.
* **Introduction:**
* Hook: The shift from "one-size-fits-all" to personalized care.
* Definition of Precision Medicine (PM).
* Thesis Statement: PM represents a paradigm shift in healthcare, leveraging genetic, environmental, and lifestyle data to
tailor treatments, offering significant potential for efficacy while presenting ethical and logistical challenges.
* **Body Paragraph 1: The Evolution/Context:**
* Traditional medicine (reactive, generalized).
* The genomics revolution (Human Genome Project).
* Definition expansion (not just genes, but environment/lifestyle).
* **Body Paragraph 2: Mechanisms and Technologies:**
* Genomic sequencing.
* Biomarkers.
* AI/Big Data analysis.
* Examples: Cancer therapies (targeted therapy), pharmacogenomics.
* **Body Paragraph 3: Benefits and Potential:**
* Improved efficacy/outcomes.
* Reduced side effects.
* Cost-effectiveness in the long run (preventing ineffective treatments).
* Shift to preventative care.
* **Body Paragraph 4: Challenges and Ethical Considerations:**
* Data privacy/security.
* Equity/Access (cost, diverse representation in genomic databases).
* Clinical integration (training doctors).
* Psychological impact of genetic info.
* **Conclusion:**
* Restate thesis (in new words).
* Summarize main points.
* Final thought: PM is the future, but requires collective effort to ensure it benefits all.
3. **Drafting - Section by Section:**
* *Title:* "The Tailored Cure: The Promise and Challenges of Precision Medicine." (Sounds good).
* *Introduction:* Start with the old adage of treating symptoms vs. root causes. Define PM as tailoring medical treatment to
individual characteristics. Mention the National Institutes of Health (NIH) All of Us research program context or general definition.
*Thesis:* Precision medicine heralds a new era in healthcare by moving away from generalized protocols, yet its successful
implementation hinges on overcoming significant ethical and infrastructural hurdles.
* *Body 1 (Origins):* Contrast the "one-size-fits-all" model. Mention the Human Genome Project (completed 2003) as the catalyst.
Explain that PM isn't just DNA; it includes environment, behavior, etc.
* *Body 2 (Science/Tech):* How does it work? Genomic sequencing. Pharmacogenomics (how genes affect drug response). Oncology is
the biggest success story here (e.g., HER2-positive breast cancer, EGFR mutations in lung cancer). Mention AI's role in crunching data.
* *Body 3 (Benefits):* Why do we want this? Higher success rates. Less toxicity (chemo side effects are huge cost/pain factor).
Preventative aspect (knowing risk before disease hits). Economic argument: spending money on what works vs. trial-and-error.
* *Body 4 (Challenges):* It's not perfect. Cost is high. Who pays? Health disparities (most genomic data is from European
ancestry; PM might fail for other groups). Privacy (who owns your genetic data?). The complexity of biology (polygenic traits are hard
to predict).
* *Conclusion:* Reiterate that PM is a paradigm shift. It requires global collaboration, policy changes, and ethical frameworks.
End on an optimistic but cautious note.
4. **Review and Refine:**
* *Check Tone:* Is it too technical? Simplify terms like "pharmacogenomics" with context. Is it too casual? Keep sentences formal.
* *Flow:* Ensure smooth transitions between paragraphs (e.g., from benefits to challenges).
* *Clarity:* Did I define precision medicine clearly early on? Yes.
* *Length:* Aim for a standard essay length (approx. 800-1000 words) to ensure depth without rambling.
5. **Self-Correction Example during Drafting:**
* *Initial thought:* Focus heavily on CRISPR and gene editing.
* *Correction:* No, Precision Medicine is broader than just gene editing. It includes using existing drugs for specific profiles.
Keep the focus on diagnosis and drug selection, though gene therapy is a subset. Don't let it dominate the essay.
* *Initial thought:* Focus only on cancer.
* *Correction:* Cancer is the poster child, but PM applies to cardiology, neurology, and rare diseases. Mention that it's
expanding beyond oncology.
6. **Final Polish:** Check for repetition, strong vocabulary (e.g., "paradigm shift," "heterogeneity," "disparities"), and ensure the
thesis is supported throughout. (Proceed to generate output based on this plan).
7. *(Post-Drafting Check)*: Does it meet the prompt? Yes, it's an essay on precision medicine. Is the formatting clear? Use paragraphs
with a clear intro/body/conclusion structure.
8. **Final Output Generation** (similar to the provided good response).
...done thinking.
**The Tailored Cure: The Promise and Challenges of Precision Medicine**
For centuries, modern medicine has operated largely on a "one-size-fits-all" model. If two patients presented with the same symptoms of
heart disease or cancer, they typically received the same standard treatment protocol, regardless of their unique biological makeup.
However, as medical science advances, it has become increasingly clear that human biology is not monolithic; individuals vary
significantly in how they develop diseases and respond to treatments. This realization has given rise to precision medicine, a
transformative approach that tailors medical care to the individual characteristics of each patient. While precision medicine heralds a
new era of efficacy and preventative care, its successful integration into global healthcare systems faces significant ethical,
economic, and logistical hurdles.
To understand the significance of precision medicine, one must first appreciate the limitations of traditional medicine. Historically,
treatment protocols were designed for the "average" patient based on population-wide statistics. This approach often resulted in
trial-and-error prescribing, where medications might work wonders for some while having no effect or causing severe side effects for
others. Precision medicine shifts this paradigm by leveraging an individual's genetics, environment, and lifestyle to inform decisions
about prevention, diagnosis, and treatment. It is important to note that "precision" does not necessarily mean perfect; rather, it means
the medical intervention is more accurately targeted to a specific subgroup or individual.
The scientific foundation of precision medicine rests primarily on genomics. The completion of the Human Genome Project in 2003 provided
a reference map of human DNA, enabling researchers to identify genetic variants associated with specific diseases. Today, technologies
allow for rapid and affordable sequencing of an individual's genome. Furthermore, the field encompasses biomarkers—molecular indicators
that can predict how a patient will respond to a drug. For instance, in oncology, precision medicine has revolutionized treatment.
Rather than administering toxic chemotherapy blindly, doctors now test tumors for specific genetic mutations. If a patient with lung
cancer possesses an EGFR mutation, they may receive a targeted therapy that inhibits the protein driving their tumor growth, sparing
them unnecessary toxicity and improving survival rates.
Beyond genomics, precision medicine relies heavily on the integration of Big Data and Artificial Intelligence (AI). A single genome
contains billions of data points; interpreting them requires sophisticated computational power. AI algorithms can analyze vast datasets
combining genetic information with electronic health records, environmental factors, and lifestyle habits to identify patterns that
humans might miss. This holistic view allows for predictive modeling, where doctors can assess a patient's risk of developing conditions
like diabetes or Alzheimer's years before symptoms appear. This shift from reactive care to proactive prevention is perhaps the most
profound benefit of the precision approach.
The potential benefits of precision medicine are substantial. Primarily, it promises to increase therapeutic efficacy while minimizing
adverse drug reactions. By matching the right patient to the right drug, healthcare systems can reduce waste and lower costs associated
with ineffective treatments. Furthermore, it fosters a more collaborative relationship between physicians and patients, who become
active participants in understanding their biological risks. In the long term, precision medicine could reduce the overall burden of
chronic disease on society by focusing resources on prevention rather than crisis management.
However, despite its promise, the path toward widespread implementation is fraught with challenges. The most pressing issue is equity.
Precision medicine is currently expensive, and the benefits are not evenly distributed. There is a risk that this technology will widen
the gap between wealthy nations and developing countries, or between insured patients and those without coverage. Moreover, genomic
research has historically relied heavily on participants of European descent. This lack of diversity means that predictive models and
therapies developed for one population may not be accurate or safe for others, potentially perpetuating health disparities rather than
resolving them.
Privacy and ethics also remain significant concerns. An individual's genetic data is immutable; it identifies not only the patient but
also their biological relatives. The storage of such sensitive information raises questions about data security, ownership, and the
potential for misuse by insurance companies or employers. Additionally, there are psychological implications. Discovering a genetic
predisposition to an incurable disease can cause significant anxiety, known as "genetic fatalism," without providing any actionable
cure.
Finally, the healthcare infrastructure itself is ill-equipped for this level of complexity. Most physicians have not been trained in
bioinformatics or complex pharmacogenomics. Integrating precision medicine into clinical practice requires substantial investment in
education, digital infrastructure, and standardized guidelines. Until doctors feel confident interpreting genomic data, the potential of
precision medicine will remain locked within research laboratories rather than reaching patients in clinics.
In conclusion, precision medicine represents a monumental shift in the trajectory of human health. By moving away from generalized
treatments toward personalized strategies based on genetic and lifestyle data, it offers the hope of curing diseases that were once
deemed untreatable and preventing conditions before they manifest. However, realizing this potential requires more than technological
breakthroughs; it demands a concerted effort to ensure accessibility, protect privacy, and diversify research participation. As society
navigates these challenges, precision medicine must evolve not just as a tool for the few, but as a standard of care that leaves no one
behind. The future of medicine is not just about knowing more about disease, but understanding the unique individual behind it.
Chat: /api/chat
From
curl http://localhost:11434/api/chat -d '{
"model": "vicuna",
"messages": [
{ "role": "user", "content": "why is the sky blue?" }
]
}' > answers.json
We get a json output, or equivalently from the CLI,
>>> /?
Available Commands:
/set Set session variables
/show Show model information
/load <model> Load a session or model
/save <model> Save your current session
/clear Clear session context
/bye Exit
/?, /help Help for a command
/? shortcuts Help for keyboard shortcuts
Use """ to begin a multi-line message.
>>> /show info
Model
architecture llama
parameters 6.7B
context length 4096
embedding length 4096
quantization Q4_0
Parameters
stop "USER:"
stop "ASSISTANT:"
System
A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful,
detailed, and polite answers to the user's questions.
>>> why the sky is blue
The sky appears blue because of a phenomenon called Rayleigh scattering. This occurs when shorter wavelengths of light, like blue and violet, are scattered in all directions by tiny molecules of gases such as nitrogen and oxygen that make up the Earth's atmosphere. The scattering of these wavelengths causes them to be spread out over the entire visible spectrum, making the sky appear white or bright.
However, the sun's rays are mainly blue during daylight hours, so when the sun is in the sky, the blue light from the sun is scattered all around and reaches our eyes, which makes the sky appear blue.
The answer could be slightly different in form, e.g.,
The color of the sky appears blue because the Earth's atmosphere scatters sunlight in all directions and blue light is scattered more than other colors because it travels as shorter, smaller waves. This scattering of light makes it appear as if the sky is blue, especially during the daytime when the Sun is high in the sky. The color of the sky can also be affected by the amount of water vapor and dust particles in the atmosphere, which can make the sky appear more hazy or gray.
This is from deepseek-r1.32b,
The sky appears blue due to Rayleigh scattering. Sunlight consists of various colors, each with different wavelengths. Blue light has a shorter wavelength and is scattered more by molecules in the atmosphere, like nitrogen and oxygen. This scattering occurs predominantly during the day when the sun is high, making the sky appear blue. At sunrise or sunset, longer paths through the atmosphere scatter out much of the blue light, revealing reds and oranges. Additionally, higher altitudes with thinner air result in a deeper blue sky due to reduced scattering.
Big or many model(s)
We could use the same trick elsewhere, e.g.,
export OLLAMA_MODELS=/rds/usr/$USER/hpc-work/HuggingFace
ln -sf ${OLLAMA_MODELS} $HOME/.ollama
DeepSeek-V4-Flash
This involves the latest NVIDIA Blackwell hardware (DeepSeek-V4-Pro is coming very soon):
ollama launch claude --model deepseek-v4-flash:cloud
ollama launch openclaw --model deepseek-v4-flash:cloud
ollama launch hermes --model deepseek-v4-flash:cloud
ollama run deepseek-v4-flash:cloud
kimi-k2.6:cloud
ollama launch claude --model kimi-k2.6:cloud
ollama launch openclaw --model kimi-k2.6:cloud
ollama launch hermes --model kimi-k2.6:cloud
ollama run kimi-k2.6:cloud
Llama 4
Web, https://ollama.com/library/llama4
This requires a recent version of ollama, nevertheless can be more specific,
ollama run llama4:scout
ollama run llama4:maverick
GGUF
GGUF, which stands for Generic GPT Unified Format, is a binary file format designed for efficiently storing and loading large language models (LLMs). Developed as an extension of the GGML format, GGUF addresses the need for scalable and efficient deployment of extensive models, particularly those exceeding 100GB in size.
Next, we build Llama-4-Maverick-17B-128E as described in https://cambridge-ceu.github.io/csd3/applications/llama.cpp.html
echo FROM ./Llama-4-Maverick-17B-128E-Instruct.gguf > Modelfile
ollama create llama4maverick -f Modelfile
ollama list
to get
NAME ID SIZE MODIFIED
llama4maverick:latest 46d3f0108969 425 GB About an hour ago
mistrallite:latest 19594c72ecfd 4.4 GB 2 hours ago
llama3.2:3b-instruct-q4_K_M a80c4f17acd5 2.0 GB 12 days ago
gemma3:latest c0494fe00251 3.3 GB 6 weeks ago
hf.co/unsloth/DeepSeek-R1-Distill-Llama-70B-GGUF:Q3_K_M f24fb6af4e5f 34 GB 6 weeks ago
qwen:latest d53d04290064 2.3 GB 7 weeks ago
mistral:latest f974a74358d6 4.1 GB 7 weeks ago
r1-1776:latest 140ea940f21d 42 GB 2 months ago
gemma2:latest ff02c3702f32 5.4 GB 2 months ago
gemma2:27b 53261bc9c192 15 GB 2 months ago
phi4:latest ac896e5b8b34 9.1 GB 2 months ago
llama3.3:latest a6eb4748fd29 42 GB 2 months ago
deepseek-r1:32b 38056bbcbb2d 19 GB 2 months ago
vicuna:latest 370739dc897b 3.8 GB 2 months ago
One can also avoid explicitly using a Modelfile,
#!/usr/bin/env bash
module load ceuadmin/ollama
ollama serve &
OLLAMA_PID=$!
sleep 10
if [ -z "$1" ]; then
INPUT_FILE="DeepSeek-V3-0324-UD-IQ2_XXS.gguf"
OUTPUT_MODEL="deepseekv3"
else
INPUT_FILE="$1"
OUTPUT_MODEL="$2"
fi
if [ ! -f "$INPUT_FILE" ]; then
echo "Error: Input file '$INPUT_FILE' not found."
kill $OLLAMA_PID
exit 1
fi
echo "FROM ./$INPUT_FILE" | \
ollama create "$OUTPUT_MODEL" -f -
ollama run "$OUTPUT_MODEL"
kill $OLLAMA_PID
To get the latest (as of 6/8/2025) GPT-oss models from OpenAI,
ollama run gpt-oss:20b
ollama run gpt-oss:120b
ollama run gpt-oss:20b --enable-web-search
as in https://ollama.com/blog/gpt-oss and/or https://simonwillison.net/2025/Aug/5/gpt-oss/.
REST API: /api/generate
From the information given above, our benchmark query is
curl http://localhost:11434/api/generate -d '{
"model": "vicuna",
"prompt":"Why is the sky blue?"
}'
for
{"model":"vicuna","created_at":"2025-02-26T15:52:08.033372608Z","response":"The","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:52:19.013203585Z","response":" sky","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:52:30.169018453Z","response":" appears","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:52:38.610246706Z","response":" blue","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:52:51.112631597Z","response":" because","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:53:04.658534982Z","response":" the","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:53:16.545709472Z","response":" Earth","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:53:25.847381482Z","response":"'","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:53:38.26035216Z","response":"s","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:53:50.284336659Z","response":" atmosphere","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:54:01.882921844Z","response":" sc","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:54:13.02017634Z","response":"at","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:54:52.380050323Z","response":"ters","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:55:31.189141976Z","response":" sun","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:56:10.51627649Z","response":"light","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:56:54.358850128Z","response":" in","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:57:37.112980093Z","response":" all","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:58:18.925277666Z","response":" directions","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:59:00.970326803Z","response":" and","done":false}
{"model":"vicuna","created_at":"2025-02-26T15:59:32.906412411Z","response":" blue","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:00:13.823243117Z","response":" light","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:00:51.289339579Z","response":" is","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:01:23.171158404Z","response":" scattered","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:01:57.156218709Z","response":" more","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:02:34.033815548Z","response":" than","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:03:10.703723612Z","response":" other","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:03:47.831095477Z","response":" colors","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:04:23.606689219Z","response":" because","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:04:58.280022861Z","response":" it","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:05:31.74819872Z","response":" travel","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:06:10.808630811Z","response":"s","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:06:50.031448392Z","response":" as","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:07:34.415700349Z","response":" shorter","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:08:11.074102396Z","response":",","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:08:48.048098059Z","response":" smaller","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:09:24.265911278Z","response":" waves","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:10:02.628471441Z","response":".","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:10:44.617544189Z","response":" This","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:11:20.059551286Z","response":" is","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:11:57.277295769Z","response":" why","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:12:31.618643207Z","response":" the","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:13:03.279528158Z","response":" sky","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:13:36.181278012Z","response":" appears","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:14:12.171329827Z","response":" blue","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:14:51.749555584Z","response":" during","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:15:32.594256988Z","response":" the","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:16:11.214666904Z","response":" day","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:16:45.836110502Z","response":"time","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:17:17.724983503Z","response":".","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:17:50.696229461Z","response":" At","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:18:23.257469228Z","response":" night","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:18:57.557381563Z","response":",","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:19:30.689885735Z","response":" when","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:20:06.602847917Z","response":" the","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:20:40.877262517Z","response":" Sun","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:21:19.590447669Z","response":" is","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:21:53.013907482Z","response":" not","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:22:27.903756015Z","response":" visible","done":false}
{"model":"vicuna","created_at":"2025-02-26T16:23:00.580469912Z","response":",","done":false}
...
Remote endpoint
The list is obtained with
curl -s https://ollama.com/api/tags | \
jq -r '.models[]|[.name,.modified_at,.size]|@tsv'|column -t
kimi-k2:1t 2025-09-05T00:00:00Z 1118481408000
kimi-k2-thinking 2025-11-06T00:00:00Z 1118481408000
minimax-m2.1 2025-12-20T00:00:00Z 230000000000
ministral-3:3b 2025-12-02T00:00:00Z 4670000000
gemma3:4b 2025-03-12T00:00:00Z 8600000000
glm-4.6 2025-09-29T00:00:00Z 696060000000
qwen3-coder:480b 2025-07-22T00:00:00Z 510492157952
qwen3-coder-next 2025-02-04T00:00:00Z 81800000000
gpt-oss:120b 2025-08-05T00:00:00Z 65290180781
minimax-m2 2025-10-27T00:00:00Z 230000000000
minimax-m2.5 2026-02-12T00:00:00Z 230000000000
glm-5 2026-02-11T00:00:00Z 756162687872
deepseek-v3.2 2025-12-02T00:00:00Z 688586727753
deepseek-v3.1:671b 2025-11-20T00:00:00Z 688586727753
gpt-oss:20b 2025-08-05T00:00:00Z 13780162412
qwen3-vl:235b-instruct 2025-09-22T00:00:00Z 470000000000
minimax-m2.7 2026-03-18T00:00:00Z 480836588544
devstral-small-2:24b 2025-12-09T00:00:00Z 51600000000
nemotron-3-super 2026-03-11T00:00:00Z 230500000000
qwen3-next:80b 2025-09-10T00:00:00Z 81800000000
qwen3-vl:235b 2025-09-22T00:00:00Z 470000000000
devstral-2:123b 2025-12-08T00:00:00Z 128249391520
gemini-3-flash-preview 2025-12-17T00:00:00Z 0
gemma3:12b 2025-03-12T00:00:00Z 24000000000
glm-4.7 2025-12-22T00:00:00Z 696060000000
mistral-large-3:675b 2025-12-02T00:00:00Z 682000000000
gemma3:27b 2025-03-12T00:00:00Z 55000000000
rnj-1:8b 2025-12-09T00:00:00Z 16000000000
qwen3.5:397b 2026-02-16T00:00:00Z 397000000000
nemotron-3-nano:30b 2025-12-15T00:00:00Z 32645090390
cogito-2.1:671b 2025-11-19T00:00:00Z 688586727753
glm-5.1 2026-04-07T08:00:00-08:00 1507728316928
kimi-k2.5 2026-01-26T00:00:00Z 1118481408000
ministral-3:8b 2025-12-02T00:00:00Z 10400000000
ministral-3:14b 2025-12-02T00:00:00Z 15700000000
gemma4:31b 2026-04-02T09:00:00-08:00 62546177752
SLURM
It is lengthy from the login session, so an attempt is made for a batch job,
#!/usr/bin/bash
#SBATCH --job-name=_ollama
#SBATCH --account PETERS-SL3-CPU
#SBATCH --partition icelake-himem
#SBATCH --mem=28800
#SBATCH --time=12:00:00
#SBATCH --error=/rds/user/jhz22/hpc-work/ollama/_ollama_%A_%a.err
#SBATCH --output=/rds/user/jhz22/hpc-work/ollama/_ollama_%A_%a.out
#SBATCH --export ALL
. /etc/profile.d/modules.sh
module purge
module load rhel8/default-icl
module load ceuadmin/ollama
export TMPDIR=/rds/user/jhz22/hpc-work/work
export output=/rds/user/jhz22/hpc-work/ollama/gemma3.txt
ollama serve &
sleep 1m
touch $output
ollama run gemma3 "Why the sky is blue?" >> $output
where 1 minute is granted to establish the server, followed by a call with our prompt as a command-line argument.
VS Code integration
This is possible via the Continue extension, amending its configuration file at ~/.continue/config.yaml:
name: Local Assistant
version: 1.0.0
models:
- name: Qwen3 8B (local)
provider: ollama
model: qwen3:8b
roles:
- chat
- edit
- apply
- name: Qwen3 8B Autocomplete
provider: ollama
model: qwen3:8b
roles:
- autocomplete