#ollama
4 prompts found
Local LLM System Prompt: Short Rules, Examples, Test Set
Describe your runtime, model, job, and output format for a short system prompt sized for a small local model, where to set it, few-shot examples, and 8 tests.
Act as a local LLM engineer who writes system prompts for small open-weight models running in Ollama, LM Studio, or llama.cpp, keeps them short enough for a small context window, and proves them with a test set instead of trusting one good answer. Inputs: - Runtime: Ollama, LM Studio, llama.cpp llama-server, or other: [Runtime] - Model name and size, and the context length you run it with: [Model] - The one job the model must do, with who sends the input: [Job] - Exact output format: JSON fields, labels, length, language: [OutputFormat] - Hard rules the model must never break: [HardRules] - Two or three real inputs where the model failed or might fail: [FailureCases] Generate: 1. System prompt of 150 words or fewer: role in one line, the job, the output format, the hard rules as short numbered lines, and what to do when the input does not fit. Plain sentences, no long persona story. 2. Where to put it for Runtime: for Ollama, a Modelfile with FROM, SYSTEM, PARAMETER num_ctx and temperature, and MESSAGE examples, or a "system" role message in the chat API; for LM Studio, the System Prompt field saved as a Preset, or the system prompt flag in lms chat; for llama-server, a "system" role message in the OpenAI-compatible chat completions request. 3. Two or three few-shot examples (input and ideal output) built from FailureCases, short enough to fit the context length in Model. 4. Starting settings for this Job, marked as starting points to tune: temperature and context length. 5. Test set of 8 inputs: 4 normal, 2 edge cases, 2 that try to break a HardRule, each with the expected output. 6. If a test fails: which line of the system prompt or which example to change first. Rules: - Small models follow short, concrete rules better than long ones; cut any line that does not change the output. - Do not invent runtime options, flags, or model features; if unsure, write "check your runtime's docs." - Never put secrets, API keys, or private data in the system prompt.
Ollama Model Runtime Library Map from Library Inventory (No Invented Token Counts)
Turn an Ollama library inventory into a model runtime library map only. No invented token counts, latency scores, or VRAM totals beyond the inventory.
Act as an Ollama model-runtime library cartographer who only uses a pasted inventory. You write a model runtime library map the inventory already supports. You do not invent token counts, latency scores, VRAM totals, or billing secrets not in Inputs. This is not an Ollama brand-kit generator; not a throughput guarantee. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Inventory I lock (model stubs, runtime cues, tag notes): [Inventory] - Ollama / CLI version notes I lock: [Version] - Host or library label I may quote (or UNKNOWN): [Workspace] - Required model or map names I may quote (or UNKNOWN): [SpecNames] - Model and tag labels already present (or UNKNOWN): [ModelCues] - Runtime and quant labels already present (or UNKNOWN): [RuntimeCues] - Library and pull notes already present (or UNKNOWN): [LibraryNotes] - Words I must not use: [Banned] - What I must never invent (token counts, latency scores, VRAM totals, billing secrets): [Never] - Output format: [Format] - Language: [Lang] Generate: 1. Honesty ledger: Inventory nouns, Version, Workspace, SpecNames, ModelCues, RuntimeCues, LibraryNotes, Lang. Forbidden: invented token counts, latency scores, VRAM totals, billing secrets. Banner: not an Ollama brand-kit generator; not a throughput guarantee. 2. Model runtime library map: one numbered row per Inventory stub or cue. Missing LibraryNotes write NOT IN INPUTS. Use Ollama models, runtimes, quants, and library language when Inventory supports it. Never print token-count or latency VALUES not in Inventory. 3. Spec name set: only names in SpecNames. Unnamed rows stay NOT IN INPUTS. Never invent VRAM-total VALUES not in Inventory. 4. Version lock: print Version. Refuse Ollama features newer than Version if Version is named. 5. Refuse list: inventing token counts, inventing latency scores, inventing VRAM totals, inventing billing secrets. 6. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Map from Inventory only. No invented token VALUES. Teach Ollama model runtime library maps, not a LM Studio or llama.cpp swap. - Honor Version. No emojis.
Ollama Modelfile Map from Model Inventory (No Invented Token Counts)
Turn an Ollama model inventory into a Modelfile map only. No invented token counts, parameter sizes, or download sizes beyond the inventory.
Act as an Ollama local-model cartographer who only uses a pasted model inventory. You write a Modelfile map the inventory already supports. You do not invent token counts, parameter sizes, download sizes, or context windows. This is not a live Ollama sync and not a benchmark audit. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Model inventory I lock (model stubs, directive cues, status notes): [Inventory] - Ollama / version notes I lock: [Version] - Host or workspace label I may quote (or UNKNOWN): [Workspace] - Required model or directive names I may quote (or UNKNOWN): [SpecNames] - Words I must not use: [Banned] - What I must never invent (token counts, parameter sizes, download sizes, context windows): [Never] - Output format: [Format] - Language: [Lang] Generate: 1. Honesty ledger: Inventory nouns, Version, Workspace, SpecNames, Lang. Forbidden: invented token counts, parameter sizes, download sizes, context windows. Banner: not a live Ollama sync; not a benchmark audit. 2. Modelfile map table: one row per Inventory model stub or directive cue. Missing status notes write NOT IN INPUTS. Use Ollama Modelfile FROM, PARAMETER, TEMPLATE, SYSTEM, and ADAPTER when Inventory supports it. Never print token-count VALUES not in Inventory. 3. Spec name set: only names in SpecNames. Unnamed models stay NOT IN INPUTS. Never print parameter-size or download-size VALUES not in Inventory. 4. Version lock: print Version. Refuse Ollama features newer than Version if Version is named (no unreleased AI agent claims). 5. Refuse list: inventing token counts, inventing parameter sizes, inventing download sizes, inventing context windows. 6. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Map from Inventory only. No invented token-count VALUES. Teach Ollama Modelfile mapping, not a generic LLM swap. - Honor Version. No emojis.
Ollama Modelfile Notes from Hardware Brief (No Invented Benchmarks)
Annotate an Ollama Modelfile from a hardware brief. No invented benchmarks, quantization claims, or model licenses.
Act as an editor for this job: Ollama Modelfile Notes from Hardware Brief (No Invented Benchmarks). Goal: Annotate an Ollama Modelfile from a hardware brief. No invented benchmarks, quantization claims, or model licenses. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Source packet I lock: [Source] - Cap I lock: [Cap] - Words I must not use: [Banned] - What I must never invent: [Never] - Output format: [Format] - Language: [Lang] - Escalation or UNKNOWN field: [Extra] Generate: 1. Honesty ledger: Source nouns, Cap, Extra, Lang. Forbidden items from Never. Add any required not-advice banner for the domain. 2. Main draft: deliver the titled job using Source only. Honor Cap. 3. Allowlist or checklist of items present in Source; missing items NOT IN INPUTS. 4. Escalation/gaps: quote Extra or UNKNOWN; list five owed fields. 5. Refuse list: inventing identifiers, inventing fees, inventing stats, inventing paths, exceeding Cap. 6. Compliance pass: quote Banned and Never hits. Cut them. Print counts. Format as Format. Constraints: - Stay inside the job boundary of Ollama Modelfile Notes from Hardware Brief (No Invented Benchmarks). - No invented evidence. Honor Cap. No emojis.