ritterkraft commited on
Commit
5cf775a
Β·
verified Β·
1 Parent(s): 97b33ee

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +43 -32
README.md CHANGED
@@ -7,70 +7,81 @@ sdk: static
7
  pinned: false
8
  short_description: Small, local models distilled from frontier teachers
9
  ---
 
10
  # Advanced Data Intelligence
 
11
  **Small, local, open models β€” distilled from frontier teachers.**
12
- ADI is a line of compact language models built at [theLAB](https://thelabsource.com)
13
- (*Learning. Algorithms. Breakthroughs.*). Each model is a **knowledge distillation**:
14
- a strong frontier "teacher" generates high-quality answers across thousands of
15
- prompts, and a small "student" model is fine-tuned to imitate them β€” producing a
16
- model that reasons and responds like something much larger, while staying small
17
- enough to run on a single consumer GPU.
18
- Every model here is built end-to-end on theLAB hardware β€” no cloud training β€” then
19
- quantized to GGUF and shipped ready to run in [Ollama](https://ollama.com) or any
20
- llama.cpp-based runtime.
21
  ---
 
22
  ## Models
 
23
  ### 🐱 adi-qwen3.5-4b-glm5.2-general
24
- General-purpose local assistant. Qwen3.5-4B distilled from **glm-5.2**.
25
- Reasons and explains like a frontier model on general topics. Native tool-calling,
26
- 262K context, ~2.7 GB.
27
  ```bash
28
  ollama run hf.co/AdvancedDataIntelligence/adi-qwen3.5-4b-glm5.2-general-GGUF:Q4_K_M
29
  ```
 
30
  ### 🐱 adi-qwen3-8b-glm5.2-general
31
- General-purpose local assistant. Qwen3-8B distilled from **glm-5.2**.
32
- Reasons and explains like a frontier model on general topics, with more headroom
33
- than the 4B. Native tool-calling, 262K context, ~5 GB.
34
  ```bash
35
  ollama run hf.co/AdvancedDataIntelligence/adi-qwen3-8b-glm5.2-general-GGUF:Q4_K_M
36
  ```
 
37
  ### 🐱 adi-qwen2.5-coder-7b-kimi2.7-code
38
- Local coding assistant. Qwen2.5-Coder-7B distilled from **kimi-k2.7-code**.
39
- Writes, explains, and debugs code with frontier-style quality. Native tool-calling,
40
- 128K context, ~4.4 GB.
41
  ```bash
42
  ollama run hf.co/AdvancedDataIntelligence/adi-qwen2.5-coder-7b-kimi2.7-code-GGUF:Q4_K_M
43
  ```
 
44
  ---
 
45
  ## How to run
46
- **Ollama (recommended).** Pull and run any model directly from this org β€” no manual
47
- download needed. Ollama fetches the GGUF from Hugging Face on first run:
 
48
  ```bash
49
  ollama run hf.co/AdvancedDataIntelligence/adi-qwen3-8b-glm5.2-general-GGUF:Q4_K_M
50
  ```
51
- Swap `:Q4_K_M` for another quant tag if a model ships multiple. To pull without
52
- running:
 
53
  ```bash
54
  ollama pull hf.co/AdvancedDataIntelligence/adi-qwen3-8b-glm5.2-general-GGUF:Q4_K_M
55
  ```
 
56
  **Manual download (llama.cpp or offline).** Grab the raw GGUF with the Hugging Face CLI:
 
57
  ```bash
58
  huggingface-cli download AdvancedDataIntelligence/adi-qwen3-8b-glm5.2-general-GGUF adi-qwen3-8b-glm5.2-q4_k_m.gguf --local-dir .
59
  ```
 
60
  Then point any llama.cpp-based runtime at the downloaded file.
 
61
  ---
 
62
  ## The approach
63
- - **Distillation, not retraining.** We transfer a teacher's reasoning style and
64
- answer quality into a small student β€” not net-new facts. For raw recall, pair
65
- these with retrieval (RAG).
66
- - **Local-first.** Every model runs fully offline on consumer hardware. No API, no
67
- data leaving the machine.
68
- - **Open.** Apache-2.0 where the base license allows, with full training details on
69
- each model card so the work is reproducible.
70
  ---
 
71
  ## Naming
72
- Models follow the pattern `adi-<base>-<size>-<teacher>-<purpose>` β€” so the name
73
- tells you the student base, its size, the teacher it learned from, and what it's
74
- tuned for.
75
  ---
 
76
  *Built at [theLAB](https://thelabsource.com) β€” Learning. Algorithms. Breakthroughs.*
 
7
  pinned: false
8
  short_description: Small, local models distilled from frontier teachers
9
  ---
10
+
11
  # Advanced Data Intelligence
12
+
13
  **Small, local, open models β€” distilled from frontier teachers.**
14
+
15
+ ADI is a line of compact language models built at [theLAB](https://thelabsource.com) (*Learning. Algorithms. Breakthroughs.*). Each model is a **knowledge distillation**: a strong frontier "teacher" generates high-quality answers across thousands of prompts, and a small "student" model is fine-tuned to imitate them β€” producing a model that reasons and responds like something much larger, while staying small enough to run on a single consumer GPU.
16
+
17
+ Every model here is built end-to-end on theLAB hardware β€” no cloud training β€” then quantized to GGUF and shipped ready to run in [Ollama](https://ollama.com) or any llama.cpp-based runtime.
18
+
 
 
 
 
19
  ---
20
+
21
  ## Models
22
+
23
  ### 🐱 adi-qwen3.5-4b-glm5.2-general
24
+
25
+ General-purpose local assistant. Qwen3.5-4B distilled from **glm-5.2**. Reasons and explains like a frontier model on general topics. Native tool-calling, 262K context, ~2.7 GB.
26
+
27
  ```bash
28
  ollama run hf.co/AdvancedDataIntelligence/adi-qwen3.5-4b-glm5.2-general-GGUF:Q4_K_M
29
  ```
30
+
31
  ### 🐱 adi-qwen3-8b-glm5.2-general
32
+
33
+ General-purpose local assistant. Qwen3-8B distilled from **glm-5.2**. Reasons and explains like a frontier model on general topics, with more headroom than the 4B. Native tool-calling, 262K context, ~5 GB.
34
+
35
  ```bash
36
  ollama run hf.co/AdvancedDataIntelligence/adi-qwen3-8b-glm5.2-general-GGUF:Q4_K_M
37
  ```
38
+
39
  ### 🐱 adi-qwen2.5-coder-7b-kimi2.7-code
40
+
41
+ Local coding assistant. Qwen2.5-Coder-7B distilled from **kimi-k2.7-code**. Writes, explains, and debugs code with frontier-style quality. Native tool-calling, 128K context, ~4.4 GB.
42
+
43
  ```bash
44
  ollama run hf.co/AdvancedDataIntelligence/adi-qwen2.5-coder-7b-kimi2.7-code-GGUF:Q4_K_M
45
  ```
46
+
47
  ---
48
+
49
  ## How to run
50
+
51
+ **Ollama (recommended).** Pull and run any model directly from this org β€” no manual download needed. Ollama fetches the GGUF from Hugging Face on first run:
52
+
53
  ```bash
54
  ollama run hf.co/AdvancedDataIntelligence/adi-qwen3-8b-glm5.2-general-GGUF:Q4_K_M
55
  ```
56
+
57
+ Swap `:Q4_K_M` for another quant tag if a model ships multiple. To pull without running:
58
+
59
  ```bash
60
  ollama pull hf.co/AdvancedDataIntelligence/adi-qwen3-8b-glm5.2-general-GGUF:Q4_K_M
61
  ```
62
+
63
  **Manual download (llama.cpp or offline).** Grab the raw GGUF with the Hugging Face CLI:
64
+
65
  ```bash
66
  huggingface-cli download AdvancedDataIntelligence/adi-qwen3-8b-glm5.2-general-GGUF adi-qwen3-8b-glm5.2-q4_k_m.gguf --local-dir .
67
  ```
68
+
69
  Then point any llama.cpp-based runtime at the downloaded file.
70
+
71
  ---
72
+
73
  ## The approach
74
+
75
+ - **Distillation, not retraining.** We transfer a teacher's reasoning style and answer quality into a small student β€” not net-new facts. For raw recall, pair these with retrieval (RAG).
76
+ - **Local-first.** Every model runs fully offline on consumer hardware. No API, no data leaving the machine.
77
+ - **Open.** Apache-2.0 where the base license allows, with full training details on each model card so the work is reproducible.
78
+
 
 
79
  ---
80
+
81
  ## Naming
82
+
83
+ Models follow the pattern `adi-<base>-<size>-<teacher>-<purpose>` β€” so the name tells you the student base, its size, the teacher it learned from, and what it's tuned for.
84
+
85
  ---
86
+
87
  *Built at [theLAB](https://thelabsource.com) β€” Learning. Algorithms. Breakthroughs.*