Instructions to use MouseWurX/28Bob-Riker with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use MouseWurX/28Bob-Riker with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf MouseWurX/28Bob-Riker:Q3_K_M # Run inference directly in the terminal: llama cli -hf MouseWurX/28Bob-Riker:Q3_K_M
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf MouseWurX/28Bob-Riker:Q3_K_M # Run inference directly in the terminal: llama cli -hf MouseWurX/28Bob-Riker:Q3_K_M
Build from source code
git clone https://github.com/ggml-org/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-app # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama serve -hf MouseWurX/28Bob-Riker:Q3_K_M # Run inference directly in the terminal: ./build/bin/llama cli -hf MouseWurX/28Bob-Riker:Q3_K_M
Use Docker
docker model run hf.co/MouseWurX/28Bob-Riker:Q3_K_M
- Llama new
- LM Studio
- Jan
- Ollama
How to use MouseWurX/28Bob-Riker with Ollama:
ollama run hf.co/MouseWurX/28Bob-Riker:Q3_K_M
- Unsloth Desktop
- Pi
How to use MouseWurX/28Bob-Riker with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf MouseWurX/28Bob-Riker:Q3_K_M
Configure the model in Pi
# Install Pi: npm install -g @earendil-works/pi-coding-agent # Add to ~/.pi/agent/models.json: { "providers": { "llama-cpp": { "baseUrl": "http://localhost:9931/v1", "api": "openai-completions", "apiKey": "none", "models": [ { "id": "MouseWurX/28Bob-Riker:Q3_K_M" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Docker Model Runner
How to use MouseWurX/28Bob-Riker with Docker Model Runner:
docker model run hf.co/MouseWurX/28Bob-Riker:Q3_K_M
- Lemonade
How to use MouseWurX/28Bob-Riker with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull MouseWurX/28Bob-Riker:Q3_K_M
Run and chat with the model
lemonade run user.28Bob-Riker-Q3_K_M
List all available models
lemonade list
- Hermes Agent
How to use MouseWurX/28Bob-Riker with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf MouseWurX/28Bob-Riker:Q3_K_M
Configure Hermes
# Install Hermes: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash hermes setup # Point Hermes at the local server: hermes config set model.provider custom hermes config set model.base_url http://127.0.0.1:9931/v1 hermes config set model.default MouseWurX/28Bob-Riker:Q3_K_M
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use MouseWurX/28Bob-Riker with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf MouseWurX/28Bob-Riker:Q3_K_M
Configure OpenClaw
# Install OpenClaw: npm install -g openclaw@latest # Register the local server and set it as the default model: openclaw onboard --non-interactive --mode local \ --auth-choice custom-api-key \ --custom-base-url http://127.0.0.1:9931/v1 \ --custom-model-id "MouseWurX/28Bob-Riker:Q3_K_M" \ --custom-provider-id llama-cpp \ --custom-compatibility openai \ --custom-text-input \ --accept-risk \ --skip-health
Run OpenClaw
openclaw agent --local --agent main --message "Hello from Hugging Face"
28Bob GGUF files, cards, keep list and adapter from Bob's Forge
Browse files- .gitattributes +2 -0
- 28Bob-Q3_K_M.bob.json +56 -57
- 28Bob-Q3_K_M.gguf +3 -0
- 28Bob-Q4_K_M.bob.json +56 -57
- 28Bob-Q4_K_M.gguf +3 -0
- README.md +25 -85
.gitattributes
CHANGED
|
@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
+
28Bob-Q3_K_M.gguf filter=lfs diff=lfs merge=lfs -text
|
| 37 |
+
28Bob-Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text
|
28Bob-Q3_K_M.bob.json
CHANGED
|
@@ -1,58 +1,57 @@
|
|
| 1 |
-
{
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
}
|
| 58 |
}
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"card": 1,
|
| 3 |
+
"file": "28Bob-Q3_K_M.gguf",
|
| 4 |
+
"bytes": 13873840224,
|
| 5 |
+
"sha256": "d3b65bbc3ae8529f721675350b2f2a6e12e3205269345eaf6066d3e535684267",
|
| 6 |
+
"name": "28Bob Riker (Q3_K_M)",
|
| 7 |
+
"nickname": "Riker",
|
| 8 |
+
"family": "Bob",
|
| 9 |
+
"madeByBob": true,
|
| 10 |
+
"jobs": [
|
| 11 |
+
"bob",
|
| 12 |
+
"deep",
|
| 13 |
+
"tinker",
|
| 14 |
+
"reviewer",
|
| 15 |
+
"office",
|
| 16 |
+
"ledger",
|
| 17 |
+
"scout"
|
| 18 |
+
],
|
| 19 |
+
"params": 29,
|
| 20 |
+
"activeParams": 3,
|
| 21 |
+
"vramGB": 15,
|
| 22 |
+
"cutFrom": {
|
| 23 |
+
"name": "Qwen3.6 35B A3B",
|
| 24 |
+
"by": "Alibaba",
|
| 25 |
+
"hf": "Qwen/Qwen3.6-35B-A3B"
|
| 26 |
+
},
|
| 27 |
+
"published": "2026-11-09",
|
| 28 |
+
"license": "Apache 2.0",
|
| 29 |
+
"what": "A Bob cut from Qwen3.6-35B-A3B for Bob's own jobs: tools, coding, exact formats and everyday chat.",
|
| 30 |
+
"benchmark": {
|
| 31 |
+
"date": "2026-11-08",
|
| 32 |
+
"report": "https://github.com/MouseWurX/Bob/blob/main/docs/FORGE-JOURNAL.md",
|
| 33 |
+
"scores": {
|
| 34 |
+
"Coding": 69,
|
| 35 |
+
"Quick code": 100,
|
| 36 |
+
"Tools": 65,
|
| 37 |
+
"Facts": 90,
|
| 38 |
+
"Routing": 96,
|
| 39 |
+
"ARC-Challenge": 92.9,
|
| 40 |
+
"ARC-Easy": 97.6,
|
| 41 |
+
"ARC-Challenge (thinking)": 93.9,
|
| 42 |
+
"ARC-Easy (thinking)": 98.3
|
| 43 |
+
},
|
| 44 |
+
"baseline": {
|
| 45 |
+
"name": "Qwen3.6 35B-A3B",
|
| 46 |
+
"scores": {
|
| 47 |
+
"Coding": 92,
|
| 48 |
+
"Quick code": 100,
|
| 49 |
+
"Tools": 67,
|
| 50 |
+
"Facts": 100,
|
| 51 |
+
"ARC-Challenge": 95.8,
|
| 52 |
+
"ARC-Easy": 98.9
|
| 53 |
+
}
|
| 54 |
+
},
|
| 55 |
+
"arcNote": "ARC zero-shot, full test sets (1,172 / 2,376); thinking on with a 1,536-token budget. Baseline thinking-off on full sets."
|
| 56 |
+
}
|
|
|
|
| 57 |
}
|
28Bob-Q3_K_M.gguf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d3b65bbc3ae8529f721675350b2f2a6e12e3205269345eaf6066d3e535684267
|
| 3 |
+
size 13873840224
|
28Bob-Q4_K_M.bob.json
CHANGED
|
@@ -1,58 +1,57 @@
|
|
| 1 |
-
{
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
}
|
| 58 |
}
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"card": 1,
|
| 3 |
+
"file": "28Bob-Q4_K_M.gguf",
|
| 4 |
+
"bytes": 17494120544,
|
| 5 |
+
"sha256": "c96ae59c62b6e1802788d07e2d3e2edb540b3a765dd71689d92a1b546e4439e7",
|
| 6 |
+
"name": "28Bob Riker (Q4_K_M)",
|
| 7 |
+
"nickname": "Riker",
|
| 8 |
+
"family": "Bob",
|
| 9 |
+
"madeByBob": true,
|
| 10 |
+
"jobs": [
|
| 11 |
+
"bob",
|
| 12 |
+
"deep",
|
| 13 |
+
"tinker",
|
| 14 |
+
"reviewer",
|
| 15 |
+
"office",
|
| 16 |
+
"ledger",
|
| 17 |
+
"scout"
|
| 18 |
+
],
|
| 19 |
+
"params": 29,
|
| 20 |
+
"activeParams": 3,
|
| 21 |
+
"vramGB": 19,
|
| 22 |
+
"cutFrom": {
|
| 23 |
+
"name": "Qwen3.6 35B A3B",
|
| 24 |
+
"by": "Alibaba",
|
| 25 |
+
"hf": "Qwen/Qwen3.6-35B-A3B"
|
| 26 |
+
},
|
| 27 |
+
"published": "2026-11-09",
|
| 28 |
+
"license": "Apache 2.0",
|
| 29 |
+
"what": "A Bob cut from Qwen3.6-35B-A3B for Bob's own jobs: tools, coding, exact formats and everyday chat.",
|
| 30 |
+
"benchmark": {
|
| 31 |
+
"date": "2026-11-08",
|
| 32 |
+
"report": "https://github.com/MouseWurX/Bob/blob/main/docs/FORGE-JOURNAL.md",
|
| 33 |
+
"scores": {
|
| 34 |
+
"Coding": 69,
|
| 35 |
+
"Quick code": 100,
|
| 36 |
+
"Tools": 65,
|
| 37 |
+
"Facts": 90,
|
| 38 |
+
"Routing": 96,
|
| 39 |
+
"ARC-Challenge": 92.9,
|
| 40 |
+
"ARC-Easy": 97.6,
|
| 41 |
+
"ARC-Challenge (thinking)": 93.9,
|
| 42 |
+
"ARC-Easy (thinking)": 98.3
|
| 43 |
+
},
|
| 44 |
+
"baseline": {
|
| 45 |
+
"name": "Qwen3.6 35B-A3B",
|
| 46 |
+
"scores": {
|
| 47 |
+
"Coding": 92,
|
| 48 |
+
"Quick code": 100,
|
| 49 |
+
"Tools": 67,
|
| 50 |
+
"Facts": 100,
|
| 51 |
+
"ARC-Challenge": 95.8,
|
| 52 |
+
"ARC-Easy": 98.9
|
| 53 |
+
}
|
| 54 |
+
},
|
| 55 |
+
"arcNote": "ARC zero-shot, full test sets (1,172 / 2,376); thinking on with a 1,536-token budget. Baseline thinking-off on full sets."
|
| 56 |
+
}
|
|
|
|
| 57 |
}
|
28Bob-Q4_K_M.gguf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c96ae59c62b6e1802788d07e2d3e2edb540b3a765dd71689d92a1b546e4439e7
|
| 3 |
+
size 17494120544
|
README.md
CHANGED
|
@@ -1,104 +1,44 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
base_model: Qwen/Qwen3.6-35B-A3B
|
| 4 |
-
tags:
|
| 5 |
-
- bob
|
| 6 |
-
- gguf
|
| 7 |
-
- mixture-of-experts
|
| 8 |
-
- expert-pruning
|
| 9 |
-
- distillation
|
| 10 |
-
- tool-use
|
| 11 |
-
language:
|
| 12 |
-
- en
|
| 13 |
---
|
| 14 |
|
| 15 |
# 28Bob Riker
|
| 16 |
|
| 17 |
-
|
| 18 |
-
assistant [Bob](https://github.com/MouseWurX/Bob). He was cut from
|
| 19 |
-
[Qwen3.6-35B-A3B](https://huggingface.co/Qwen/Qwen3.6-35B-A3B) by keeping, in every layer, the 208 of 256 experts that
|
| 20 |
-
Bob's own jobs use. Then he was healed with a LoRA for one full pass against the full model's next-word odds, on the
|
| 21 |
-
owner's lesson set ([bob-lessons](https://huggingface.co/datasets/MouseWurX/bob-lessons), about 9,000 of Bob's
|
| 22 |
-
conversations) plus general text. He was finished on 7 November 2026. Method, numbers and logs: the
|
| 23 |
-
[Forge journal](https://github.com/MouseWurX/Bob/blob/main/docs/FORGE-JOURNAL.md).
|
| 24 |
|
| 25 |
-
|
| 26 |
-
|
|
|
|
| 27 |
|
| 28 |
## Files
|
| 29 |
|
| 30 |
-
| File | Size |
|
| 31 |
|---|---|---|
|
| 32 |
-
| `28Bob-
|
| 33 |
-
| `28Bob-
|
| 34 |
-
| `28Bob-adapter.pt` | 1.7 GB | the healing LoRA; with the keep list it rebuilds the full-precision model from Qwen3.6-35B-A3B |
|
| 35 |
-
| `28Bob-keep.json` | 38 KB | the 208 experts kept in each of the 40 layers |
|
| 36 |
-
| `coverage-208.md` | small | how the keep list was chosen, and each skill's kept share |
|
| 37 |
-
| `heal-report.json` | small | the heal settings and before/after losses |
|
| 38 |
-
| `28Bob-Q4_K_M.bob.json`, `28Bob-Q3_K_M.bob.json` | small | Bob's brain cards |
|
| 39 |
|
| 40 |
-
|
| 41 |
|
| 42 |
-
|
| 43 |
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
With [llama.cpp](https://github.com/ggml-org/llama.cpp) (any recent build with Qwen3.6 MoE support):
|
| 47 |
-
|
| 48 |
-
```
|
| 49 |
-
llama-server -m 28Bob-Q4_K_M.gguf -c 32768 --jinja -fa on -ctk q8_0 -ctv q8_0 --fit on
|
| 50 |
-
```
|
| 51 |
-
|
| 52 |
-
- 8 experts per word (the file's default; don't lower it).
|
| 53 |
-
- 32K context or more. The needle test below reads cleanly at 128K.
|
| 54 |
-
- Thinking on or off per request, with the Qwen3.6 chat template (`enable_thinking`).
|
| 55 |
-
- On a 16 GB card, `--fit on` (or `--n-cpu-moe`) puts part of the experts on system memory.
|
| 56 |
-
|
| 57 |
-
## Heal (6 to 7 November 2026, one H200)
|
| 58 |
-
|
| 59 |
-
| | Before | After | Full Qwen |
|
| 60 |
-
|---|---|---|---|
|
| 61 |
-
| Held-out lessons, cross-entropy | 0.401 | 0.362 | 0.324 |
|
| 62 |
-
| Held-out lessons, KD (distance from the full model) | 0.087 | 0.067 | |
|
| 63 |
-
| General text, cross-entropy | 2.154 | 2.158 | 2.101 |
|
| 64 |
-
| General text, KD | 0.178 | 0.132 | |
|
| 65 |
-
|
| 66 |
-
929 of 929 steps, 5.9 hours, 63.4 million tokens. LoRA rank 32 (8 on the experts), learning rate 1e-4, one epoch.
|
| 67 |
-
|
| 68 |
-
## Benchmarks (8 November 2026, Q4_K_M, 8 experts per word, the owner's 16 GB card)
|
| 69 |
-
|
| 70 |
-
| Test | Riker | Full Qwen3.6-35B-A3B (Q4) |
|
| 71 |
|---|---|---|
|
| 72 |
-
|
|
| 73 |
-
|
|
| 74 |
-
|
|
| 75 |
-
|
|
| 76 |
-
|
|
| 77 |
-
|
|
| 78 |
-
|
|
| 79 |
-
|
|
| 80 |
-
|
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
ARC is zero-shot: the question and lettered choices, then "Answer with the letter only.", scored on the first letter.
|
| 85 |
-
Thinking off uses temperature 0. Thinking on gets a 1,536-token thinking budget and is then asked to answer.
|
| 86 |
-
|
| 87 |
-
The other tests are the project's own, run on one machine, for comparing Bobs with each other and with the model they
|
| 88 |
-
came from. They are not leaderboard numbers.
|
| 89 |
-
|
| 90 |
-
## Honest notes
|
| 91 |
-
|
| 92 |
-
- Facts: 18 of 20 where the full model gets 20. Trivia recall is the main thing the cut cost.
|
| 93 |
-
- The keep list was built with the plain 95% coverage rule for every skill. The stronger 99% rule for tool calls,
|
| 94 |
-
coding and exact formats, planned for later Bobs, was not used for Riker. Harder coding is where he trails most
|
| 95 |
-
(9 of 13 against 12).
|
| 96 |
-
- The heal used the owner's own lesson set, so Riker leans towards Bob's jobs: tool calls, routing, coding help and
|
| 97 |
-
office tasks.
|
| 98 |
|
| 99 |
## Licence and credit
|
| 100 |
|
| 101 |
-
Apache 2.0,
|
| 102 |
-
|
| 103 |
-
the
|
| 104 |
-
Alibaba.
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
base_model: Qwen/Qwen3.6-35B-A3B
|
| 4 |
+
tags: [bob, gguf, mixture-of-experts, tool-use]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
---
|
| 6 |
|
| 7 |
# 28Bob Riker
|
| 8 |
|
| 9 |
+
A Bob cut from Qwen3.6-35B-A3B for Bob's own jobs: tools, coding, exact formats and everyday chat.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
|
| 11 |
+
Made by [Bob's Forge](https://github.com/MouseWurX/Bob/blob/main/docs/FORGE-JOURNAL.md) from [Qwen/Qwen3.6-35B-A3B](https://huggingface.co/Qwen/Qwen3.6-35B-A3B)
|
| 12 |
+
(all experts kept per layer, about 29 B parameters, 3 B active per word).
|
| 13 |
+
Method, numbers and logs are in the Forge journal.
|
| 14 |
|
| 15 |
## Files
|
| 16 |
|
| 17 |
+
| File | Size | Needs |
|
| 18 |
|---|---|---|
|
| 19 |
+
| `28Bob-Q3_K_M.gguf` | 13.9 GB | about 15 GB of card memory |
|
| 20 |
+
| `28Bob-Q4_K_M.gguf` | 17.5 GB | about 19 GB of card memory |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
|
| 22 |
+
Each GGUF has a `.bob.json` card beside it that Bob's catalogue reads when you add the file as your own brain.
|
| 23 |
|
| 24 |
+
## Home benchmarks (2026-11-08)
|
| 25 |
|
| 26 |
+
| Test | 28Bob | Qwen3.6 35B-A3B |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 27 |
|---|---|---|
|
| 28 |
+
| Coding | 69 | 92 |
|
| 29 |
+
| Quick code | 100 | 100 |
|
| 30 |
+
| Tools | 65 | 67 |
|
| 31 |
+
| Facts | 90 | 100 |
|
| 32 |
+
| Routing | 96 | |
|
| 33 |
+
| ARC-Challenge | 92.9 | 95.8 |
|
| 34 |
+
| ARC-Easy | 97.6 | 98.9 |
|
| 35 |
+
| ARC-Challenge (thinking) | 93.9 | |
|
| 36 |
+
| ARC-Easy (thinking) | 98.3 | |
|
| 37 |
+
|
| 38 |
+
The project's own tests on one machine, for comparing Bobs with each other and with the model they came from, not for leaderboards.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 39 |
|
| 40 |
## Licence and credit
|
| 41 |
|
| 42 |
+
Apache 2.0, as the base model (Alibaba's Qwen team). The lessons it learned are Bob's own lesson sets (CC BY-SA 4.0) plus
|
| 43 |
+
openly licensed traces from other models, credited in the Forge journal. The Bob names are a tribute to Dennis E. Taylor's novels;
|
| 44 |
+
the project is not affiliated with the author.
|
|
|