Exploring 100M Tokens with DeepSeek‑Harness: A Hands‑On Performance Guide
The author runs a deepseek‑harness session to analyze its source code and build a custom skin, reporting smooth operation, fast token processing (98.3 M input, 360 K output, 131 tok/s), high cache hit rate, and detailing installation, usage, and encountered quirks.
Performance Summary
36 rounds, 348 steps executed. LLM runtime 58 m 48 s, tool calls 36 m 47 s. First token average 2.2 s → 131 tokens/s. Cache hit rate 99 %. Processed 98.3 M input tokens, generated 360 K output tokens.
Source Code Review & Project Implementation
Completed source code analysis and created a new project using a grill‑style Skill without major errors. Bug fixing and localization were accurate.
Observed Issues
File edit/write errors when a file is written without being read first.
After a long web session, the input box becomes noticeably laggy; one occurrence caused a page crash.
Skill commands that are ambiguous (e.g., “/xxx”) are sometimes interpreted as explanations instead of execution.
Installation
npx @deepseek-ai/dsh webRequires Node ≥ 22.19.0. Add an API key from https://platform.deepseek.com/api_keys.
UI Walkthrough
Steps demonstrated with screenshots:
Select working directory.
Configure permissions.
Use the tool and view logs.
Common commands.
Enable a sub‑agent.
Apply Skills (grill series installed locally).
Generate an image (yellow duck riding a bicycle).
All components function as plugins.
Skin Plugin Creation
Creation of a skin plugin shown with screenshots, culminating in the final deepseek‑harness skin view.
Repositories
Source code: https://github.com/deepseek-ai/deepseek-harness
Skin repository (npm not yet published): https://github.com/goodpostidea-tech/deepseek-harness-skin
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
AI Engineer Programming
In the AI era, defining problems is often more important than solving them; here we explore AI's contradictions, boundaries, and possibilities.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.