Hugging Face's logo

210Codelabs
/
AdaptIQ-7B

Text Generation
Transformers
adaptiq
210-code-labs
code
agent
routed-experts
openai-compatible
Model card Files
xet
 Community
210Codelabs commited on
Commit
b93c2dc
·
verified ·
1 Parent(s): fc196ef

Publish AdaptIQ-7B artifacts

Browse files
Files changed (1) hide show
  1. README.md +65 -7
README.md CHANGED
@@ -1,20 +1,78 @@
1
  ---
2
  library_name: transformers
3
  pipeline_tag: text-generation
 
4
  tags:
5
  - adaptiq
6
  - 210-code-labs
7
  - code
8
  - agent
9
- - merged
 
10
  ---
11
 
12
  # AdaptIQ-7B
13
 
14
- AdaptIQ expert composition using **mistralai/Mistral-7B-Instruct-v0.3** for text, **ibm-granite/granite-8b-code-instruct** for code,
15
- and **black-forest-labs/FLUX.1-schnell** for image generation. Composition method: `routed_experts`
16
- (alpha=0.5 is used only for compatible linear merge builds).
17
 
18
- Capabilities: text chat, code generation, agent workflows, and image generation.
19
- Image generation is provided by the companion image expert at inference time
20
- through the AdaptIQ router.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  library_name: transformers
3
  pipeline_tag: text-generation
4
+ license: apache-2.0
5
  tags:
6
  - adaptiq
7
  - 210-code-labs
8
  - code
9
  - agent
10
+ - routed-experts
11
+ - openai-compatible
12
  ---
13
 
14
  # AdaptIQ-7B
15
 
16
+ AdaptIQ is an open-weight, self-hosted routed-expert AI system from **210 Code Labs**.
17
+ This repository publishes the AdaptIQ model card, manifest, and deployment metadata for
18
+ `AdaptIQ-7B`.
19
 
20
+ ## Architecture
21
+
22
+ AdaptIQ uses expert composition instead of unsafe direct weight blending across incompatible
23
+ architectures:
24
+
25
+ | Capability | Expert |
26
+ |---|---|
27
+ | Text / reasoning | `mistralai/Mistral-7B-Instruct-v0.3` |
28
+ | Code / debugging | `ibm-granite/granite-8b-code-instruct` |
29
+ | Image generation | `black-forest-labs/FLUX.1-schnell` |
30
+
31
+ Composition method: `routed_experts`. `alpha=0.5` is retained only for compatible linear-merge
32
+ experiments and is not used by the routed-expert release path.
33
+
34
+ ## Capabilities
35
+
36
+ - General text chat
37
+ - Code generation and debugging
38
+ - Agent-oriented reasoning workflows
39
+ - Image generation through the routed image expert
40
+ - OpenAI-compatible hosted API on Modal
41
+ - MCP-ready tool schema scaffold
42
+
43
+ ## Hosted API
44
+
45
+ Modal endpoint:
46
+
47
+ ```text
48
+ https://codelabs--adaptiq-api-api.modal.run/v1
49
+ ```
50
+
51
+ Model name:
52
+
53
+ ```text
54
+ AdaptIQ-7B
55
+ ```
56
+
57
+ Example:
58
+
59
+ ```bash
60
+ curl https://codelabs--adaptiq-api-api.modal.run/v1/models \
61
+ -H "Authorization: Bearer $ADAPTIQ_API_KEY"
62
+ ```
63
+
64
+ ## Intended use
65
+
66
+ AdaptIQ is intended for software development, automation, research, education, and creative
67
+ self-hosted generation workflows.
68
+
69
+ ## Limitations
70
+
71
+ This release is a routed expert system metadata release. It does not package all upstream expert
72
+ weights into a single checkpoint. Runtime deployments download/load the configured open-weight
73
+ experts according to the AdaptIQ manifest.
74
+
75
+ ## License
76
+
77
+ The AdaptIQ release metadata is Apache-2.0. Respect the licenses and model cards of all upstream
78
+ experts before redistribution or commercial deployment.