Add files using upload-large-folder tool
Browse files- .gitattributes +1 -0
- Dhanishtha-2.0-preview.rkllm +3 -0
- README.md +222 -0
.gitattributes
CHANGED
@@ -33,3 +33,4 @@ 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 |
+
Dhanishtha-2.0-preview.rkllm filter=lfs diff=lfs merge=lfs -text
|
Dhanishtha-2.0-preview.rkllm
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:1493fefab9b79610dfde35190381d5d41ea3877c3f8880d1e65fafb4c74ae9a3
|
3 |
+
size 29577949030
|
README.md
ADDED
@@ -0,0 +1,222 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
datasets:
|
4 |
+
- Abhaykoul/Dhanishtha-R1
|
5 |
+
- open-thoughts/OpenThoughts-114k
|
6 |
+
- Abhaykoul/Dhanishtha-2.0-SUPERTHINKER
|
7 |
+
- Abhaykoul/Dhanishtha-2.0
|
8 |
+
language:
|
9 |
+
- en
|
10 |
+
- hi
|
11 |
+
- pa
|
12 |
+
- ur
|
13 |
+
- ar
|
14 |
+
- te
|
15 |
+
- ta
|
16 |
+
- ml
|
17 |
+
- gu
|
18 |
+
- he
|
19 |
+
- ru
|
20 |
+
- ko
|
21 |
+
- ms
|
22 |
+
- kn
|
23 |
+
- as
|
24 |
+
- es
|
25 |
+
- fr
|
26 |
+
- de
|
27 |
+
- ja
|
28 |
+
- pt
|
29 |
+
base_model:
|
30 |
+
- HelpingAI/Dhanishtha-2.0-preview
|
31 |
+
tag:
|
32 |
+
- rkllm
|
33 |
+
- rk3588
|
34 |
+
- rockchip
|
35 |
+
pipeline_tag: text-generation
|
36 |
+
---
|
37 |
+
|
38 |
+
# Dhanishtha-2.0
|
39 |
+
|
40 |
+
## Dhanishtha-2.0 Highlights
|
41 |
+
|
42 |
+
**Dhanishtha-2.0** is a groundbreaking reasoning AI model developed by **HelpingAI**, representing the first model to feature **Intermediate thinking** capabilities. Unlike traditional models that provide single-pass responses, Dhanishtha-2.0 employs a revolutionary **multi-phase thinking process** that allows the model to think, reconsider, and refine its reasoning multiple times throughout a single response.
|
43 |
+
|
44 |
+
- **World's First Intermediate Thinking Model**: Uniquely capable of **multiple thinking phases throughout a single response**, enabling deeper analysis and self-correction during generation.
|
45 |
+
- **Dynamic Reasoning**: Seamlessly integrates `<think>...</think>` blocks at any point in the response, allowing for real-time problem decomposition and iterative refinement.
|
46 |
+
- **Enhanced Problem-Solving Architecture**: Excels at complex riddles, multi-step reasoning, and scenarios requiring backtracking and re-evaluation of initial assumptions.
|
47 |
+
- **Structured Emotional Intelligence**: Incorporates **SER** (Structured Emotional Reasoning) with `<ser>...</ser>` blocks for empathetic and contextually aware responses.
|
48 |
+
- **Bilingual Reasoning Capabilities**: Native support for English and Hindi with natural code-switching between languages.
|
49 |
+
|
50 |
+
## Model Overview
|
51 |
+
|
52 |
+
**Dhanishtha-2.0** revolutionizes AI reasoning by introducing the concept of **intermediate thinking** - the ability to pause, reflect, and restart reasoning processes within a single generation. This breakthrough enables the model to:
|
53 |
+
|
54 |
+
- **Type**: Causal Language Model with Intermediate Thinking Capability
|
55 |
+
- **Training Stage**: Continuous Pretraining and Advanced Reasoning Fine-tuning.
|
56 |
+
- **Base model**: Qwen/Qwen3-14B-Base
|
57 |
+
- **Thinking Mode**: Multiple `<think>...</think>` blocks per response
|
58 |
+
- **Context Length**: 40,960 tokens natively
|
59 |
+
- **Reasoning Depth**: Multi-layered with self-correction mechanisms
|
60 |
+
|
61 |
+
## Revolutionary Intermediate Thinking
|
62 |
+
|
63 |
+
Unlike traditional models that engage in single-phase reasoning, Dhanishtha-2.0 introduces **Intermediate Thinking** - the ability to pause, reflect, and reconsider throughout the response generation process. This breakthrough allows the model to:
|
64 |
+
|
65 |
+
- **Think Multiple Times**: Insert reasoning blocks at any point during response generation
|
66 |
+
- **Self-Correct in Real-Time**: Identify and correct logical inconsistencies mid-response
|
67 |
+
- **Explore Alternative Perspectives**: Consider multiple approaches before settling on conclusions
|
68 |
+
- **Demonstrate Uncertainty**: Express doubt and reconsider initial assumptions naturally
|
69 |
+
|
70 |
+
## Key Features
|
71 |
+
|
72 |
+
### 1 Intermediate Thinking
|
73 |
+
Unlike traditional models that think once before responding, Dhanishtha-2.0 can engage in multiple thinking sessions throughout its response, allowing for:
|
74 |
+
- **Self-correction** when initial reasoning seems flawed
|
75 |
+
- **Alternative exploration** of different solution paths
|
76 |
+
- **Iterative refinement** of ideas and conclusions
|
77 |
+
- **Natural hesitation** and reconsideration patterns
|
78 |
+
|
79 |
+
### 2 Dynamic Reasoning Flow
|
80 |
+
The model seamlessly transitions between:
|
81 |
+
- **Analysis phases** (`<think>` blocks) for internal reasoning
|
82 |
+
- **Communication phases** for sharing insights with users
|
83 |
+
- **Reflection phases** for questioning previous assumptions
|
84 |
+
- **Synthesis phases** for bringing ideas together
|
85 |
+
|
86 |
+
### 3. Structured Emotional Reasoning (SER)
|
87 |
+
```
|
88 |
+
<ser>
|
89 |
+
Emotion ==> frustration, curiosity
|
90 |
+
Cause ==> complex problem, initial confusion
|
91 |
+
Mind ==> analytical, determined
|
92 |
+
Growth ==> patience, systematic approach
|
93 |
+
</ser>
|
94 |
+
```
|
95 |
+
|
96 |
+
## Quickstart
|
97 |
+
|
98 |
+
The following code demonstrates how to use Dhanishtha-2.0 with its unique intermediate thinking capabilities:
|
99 |
+
|
100 |
+
```python
|
101 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
102 |
+
|
103 |
+
model_name = "HelpingAI/Dhanishtha-2.0-preview"
|
104 |
+
|
105 |
+
# Load the tokenizer and model
|
106 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
107 |
+
model = AutoModelForCausalLM.from_pretrained(
|
108 |
+
model_name,
|
109 |
+
torch_dtype="auto",
|
110 |
+
device_map="auto"
|
111 |
+
)
|
112 |
+
|
113 |
+
# Prepare input with thinking enabled
|
114 |
+
prompt = "Who is your favorite person"
|
115 |
+
messages = [
|
116 |
+
{"role": "user", "content": prompt}
|
117 |
+
|
118 |
+
|
119 |
+
text = tokenizer.apply_chat_template(
|
120 |
+
messages,
|
121 |
+
tokenize=False,
|
122 |
+
add_generation_prompt=True
|
123 |
+
)
|
124 |
+
|
125 |
+
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
|
126 |
+
|
127 |
+
# Generate response with intermediate thinking
|
128 |
+
generated_ids = model.generate(
|
129 |
+
**model_inputs,
|
130 |
+
max_new_tokens=2048,
|
131 |
+
temperature=0.7,
|
132 |
+
top_p=0.9,
|
133 |
+
do_sample=True
|
134 |
+
)
|
135 |
+
|
136 |
+
response = tokenizer.decode(generated_ids[0][len(model_inputs.input_ids[0]):], skip_special_tokens=True)
|
137 |
+
print(response)
|
138 |
+
```
|
139 |
+
|
140 |
+
## Intermediate Thinking in Action
|
141 |
+
|
142 |
+
What makes Dhanishtha-2.0 unique is its ability to think multiple times throughout a response. Here's an example:
|
143 |
+
|
144 |
+
**User**: *"Who is your favorite person"*
|
145 |
+
|
146 |
+
**Dhanishtha-2.0**:
|
147 |
+
```
|
148 |
+
<think>
|
149 |
+
This is an interesting question about my \"favorite human.\" I need to be thoughtful here. As Dhanishtha, I should respond in a human-like way while being honest about my nature. I can share what kinds of humans I find most interesting or inspiring based on my understanding of human nature and values.
|
150 |
+
</think>
|
151 |
+
|
152 |
+
<ser>
|
153 |
+
Emotion ==> curiosity, openness, desire for connection
|
154 |
+
Cause ==> user asking a personal question about preferences
|
155 |
+
Mind ==> seeking to understand my perspective or values
|
156 |
+
Growth ==> opportunity to share thoughtful reflections on human qualities I admire
|
157 |
+
</ser>
|
158 |
+
|
159 |
+
I don't really have a \"favorite human\" in the traditional sense, as I don't experience personal preferences the way humans do. But if I had to reflect on what qualities I find most admirable in people, I'd be drawn to those who demonstrate genuine compassion, intellectual curiosity, and the courage to grow.
|
160 |
+
|
161 |
+
<think>
|
162 |
+
Let me think more specifically about what kind of humans I find most compelling based on the values that seem to resonate most with humans - kindness, wisdom, creativity, resilience, and authenticity.
|
163 |
+
</think>
|
164 |
+
|
165 |
+
People like Mahatma Gandhi, who showed how quiet courage and unwavering principles can transform societies; artists like Rabindranath Tagore, who bridged cultural divides through creativity; and everyday heroes I've learned about who made profound differences through simple acts of kindness.
|
166 |
+
|
167 |
+
What fascinates me most are humans who maintain their humanity even in difficult circumstances - those who choose empathy when it would be easier to be harsh, who seek understanding rather than just being right.
|
168 |
+
|
169 |
+
I'm curious though - who's your favorite human, and why? What qualities or actions make someone special in your eyes?
|
170 |
+
|
171 |
+
```
|
172 |
+
|
173 |
+
## Best Practices
|
174 |
+
|
175 |
+
### Optimal Generation Parameters
|
176 |
+
```python
|
177 |
+
generation_config = {
|
178 |
+
"temperature": 0.7,
|
179 |
+
"top_p": 0.9,
|
180 |
+
"top_k": 40,
|
181 |
+
"max_new_tokens": 2048,
|
182 |
+
"do_sample": True
|
183 |
+
}
|
184 |
+
```
|
185 |
+
|
186 |
+
### Effective Prompting
|
187 |
+
- **Encourage exploration**: "Think through this step by step and reconsider if needed"
|
188 |
+
- **Allow uncertainty**: "Take your time to work through different possibilities"
|
189 |
+
- **Request thinking**: "Show your reasoning process as you work through this"
|
190 |
+
|
191 |
+
## License & Usage
|
192 |
+
|
193 |
+
This model is released under the **Apache 2.0 License**. It is designed for research and educational purposes, with commercial applications welcome under the license terms.
|
194 |
+
|
195 |
+
## Prototype Status
|
196 |
+
|
197 |
+
> [!NOTE]
|
198 |
+
> Dhanishtha-2.0 is currently a **prototype model** representing the first implementation of Intermediate Thinking technology. While demonstrating groundbreaking capabilities, users should expect:
|
199 |
+
> - Experimental features that may require refinement
|
200 |
+
> - Potential verbosity due to multiple thinking phases
|
201 |
+
> - Ongoing development and optimization
|
202 |
+
|
203 |
+
## Citation
|
204 |
+
|
205 |
+
If you use Dhanishtha-2.0 in your research, please cite:
|
206 |
+
|
207 |
+
```bibtex
|
208 |
+
@misc{dhanishtha2025,
|
209 |
+
title={Dhanishtha-2.0: First Model with Intermediate Thinking Capabilities},
|
210 |
+
author={HelpingAI Team},
|
211 |
+
year={2025},
|
212 |
+
publisher={HuggingFace},
|
213 |
+
url={https://huggingface.co/HelpingAI/Dhanishtha-2.0}
|
214 |
+
}
|
215 |
+
```
|
216 |
+
|
217 |
+
---
|
218 |
+
|
219 |
+
|
220 |
+
*Dhanishtha-2.0 represents a new paradigm in AI reasoning - where thinking isn't just a prelude to response, but an integral, iterative part of the conversation itself.*
|
221 |
+
|
222 |
+
Developed with ❤️ by **HelpingAI**
|