#56

Author: miku

content/posts/meetup-56-wrapup.md

@@ -1,20 +1,23 @@
 ---
 title: "Hybrid Meetup #56 wrap-up"
 date: 2025-11-26T00:30:00+01:00
-draft: true
+draft: false
 tags:
 - summary
 - meetup
 ---
 
-### An older version of the matrix
+### An older version of the matrix / 不気味の谷
 
 Hybrid Meetup #56 took place
-[2025-11-25](https://www.meetup.com/de-de/leipzig-golang/events/305626275/) 19:00 at
-[Basislager Leipzig](https://basislager.co) and we looked into basic agents with Go.
+[2025-11-25](https://www.meetup.com/de-de/leipzig-golang/events/305626275/)
+19:00 at [Basislager Leipzig](https://basislager.co) and we looked into basic
+agents with Go, notes can be found here:
+[miku/unplugged](https://github.com/miku/unplugged).
 
-Agents are possible because of the reasoning and tool support of
-language models.
+Agents are possible because of the *reasoning* and tool support of language
+models (and they are [simple to
+write](https://fly.io/blog/everyone-write-an-agent/).
 
 An early paper on tools was [*Toolformer: Language Models Can Teach Themselves to Use Tools*](https://arxiv.org/pdf/2302.04761) (2023-02-09)
 
@@ -30,15 +33,114 @@ An early paper on tools was [*Toolformer: Language Models Can Teach Themselves t
 > Given just a handful of human-written examples of how an API can be used, we
 > let a LM annotate a huge language modeling dataset with potential API calls.
 > We then use a self-supervised loss to determine which of these API calls
-> actually help the model in predicting future tokens. Finally, we finetune the
-> LM itself on the API calls that it con- siders useful. As illustrated in
-> Figure 1, through this simple approach, LMs can learn to control a va- riety
-> of tools, and to choose for themselves which tool to use when and how.
+> actually help the model in predicting future tokens.
+
+An agentic setup then is mostly a loop that manages a context over time with
+the help of tools.
+
+![](/images/lgo-56-looptool.gif)
+
+Google Agent SDK at this time only supports gemini out of the box, so we wrote
+a simple agent from scratch (against any openai compatible endpoint) and ended
+up with an agent that had a list of tools (some of them not fully implemented):
+
+```
+get_weather
+add_numbers
+get_time
+search_library_catalog
+ping
+list_files
+read_file
+grep
+write_file
+append_file
+run_command
+```
+
+We used both an [RTX 4000 SFF
+ADA](https://www.nvidia.com/content/dam/en-zz/Solutions/rtx-4000-sff/proviz-rtx-4000-sff-ada-datasheet-2616456-web.pdf)
+and an [AMD AI MAX+ 395 with an
+8060S](https://www.amd.com/en/products/processors/laptop/ryzen/ai-300-series/amd-ryzen-ai-max-plus-395.html).
+
+| Spec | AMD Radeon 8060S | NVIDIA RTX 4000 SFF Ada |
+|------|------------------|-------------------------|
+| **FP16 (theoretical)** | **59.4 TFLOPS** | **~19.2 TFLOPS** |
+| Memory Bandwidth | ~212 GB/s (DDR5-8000) | 280 GB/s (GDDR6) |
+
+However, prefill is a bit faster on the nvidia card:
+
+```
+$ time OLLAMA_MODEL=qwen3:14b OLLAMA_HOST=http://ada:11434 ./one -m "how warm is it in leipzig?"
+2025/11/26 09:57:17 user: how warm is it in leipzig?               ...
+2025/11/26 09:57:17 context length: 4974                           ...
+2025/11/26 09:57:23 assistant wants to call 1 tool(s)              ...
+2025/11/26 09:57:23 calling tool: get_weather                      ...
+2025/11/26 09:57:23 args: {"city":"Leipzig"}                       ...
+2025/11/26 09:57:23     Result: {"city": "Leipzig", "temperature": ...
+2025/11/26 09:57:23 context length: 5227                           ...
+2025/11/26 09:57:30 assistant: The current temperature in Leipzig i...
+
+real    0m13.423s
+user    0m0.001s
+sys     0m0.013s
+
+$ time OLLAMA_MODEL=qwen3:14b OLLAMA_HOST=http://strix:11434 ./one -m "how warm is it in leipzig?"
+2025/11/26 09:57:34 user: how warm is it in leipzig?             ...
+2025/11/26 09:57:34 context length: 4974
+2025/11/26 09:57:41 assistant wants to call 1 tool(s)
+2025/11/26 09:57:41 calling tool: get_weather
+2025/11/26 09:57:41 args: {"city":"Leipzig"}
+2025/11/26 09:57:41     Result: {"city": "Leipzig", "temperature"...
+2025/11/26 09:57:41 context length: 5227
+2025/11/26 09:57:50 assistant: The current temperature in Leipzig...
+
+real    0m15.826s
+user    0m0.007s
+sys     0m0.007s
+```
+
+Still, the interplay is interesting to observe. Requests like *"save the
+temperature in leipzig to temp.txt"* or *"fetch
+[https://golangleipzig.space/leipzig-gopher.png](https://golangleipzig.space/leipzig-gopher.png)
+and convert it to jpg"* work with a 9.3GB 14B tool supporting LLM like
+[Qwen3-14B](https://huggingface.co/Qwen/Qwen3-14B).
+
+> Expertise in agent capabilities, enabling precise integration with external
+> tools in both thinking and unthinking modes and achieving leading performance
+> among open-source models in complex agent-based tasks. -- [model card](https://huggingface.co/Qwen/Qwen3-14B)
+
+Example: (1) fetch image, (2) convert to jpg, (3) calculate sha1 and (4) write the result to a file (speedup 1.5x):
+
+![](/images/lgo-56-agent-3.gif)
+
+Future ideas for tools:
+
+* [ ] browser use to facility web search
+* [ ] manage context by offloading to text files, similar to deep research agents
+* [ ] code snippets for subtasks and sandboxed execution
+
+Go's concurrency facilities seems to be helpful when implementing agents.
 
 ### Misc
 
+* There is no shortage of agent frameworks, e.g.
+  [tRPC-Agent-Go](https://github.com/trpc-group/trpc-agent-go),
+[eino](https://github.com/cloudwego/eino),
+[genkit](https://genkit.dev/docs/get-started/?lang=go),
+[swarmgo](https://github.com/prathyushnallamothu/swarmgo),
+[go-agent](https://github.com/vitalii-honchar/go-agent),
+[suricata](https://github.com/ostafen/suricata),
+[agent-sdk-go](https://github.com/Ingenimax/agent-sdk-go), and many more in Go
+or [any language](https://www.shakudo.io/blog/top-9-ai-agent-frameworks)
+* [KI BARCAMP HALLE (SAALE) 2025-11-29](https://www.klaustor-coworking.de/events-1/ki-barcamp-2025)
 * [go4lage gemini CV](https://go4lage.com/geminicv), for *escaping vendor lock-in*
+* [Can Google's ADK Replace LangChain and MCP? (with Christina Lin)](https://www.youtube.com/watch?v=nMnQ63YkftE)
+* AI will be similar to the internet, e.g. in terms of omnipresence
+* AI [uncanny valley](https://en.wikipedia.org/wiki/Uncanny_valley), 不気味の谷
 
+Did you implement a cool agent in Go (or something else)? Then why not [join
+us](https://www.meetup.com/de-DE/leipzig-golang/) and let it introduce itself.
 
 ----