Image-Text-to-Text
Transformers
Safetensors
English
llava
conversational
patricebechard commited on
Commit
e6f9254
·
verified ·
1 Parent(s): 6d4291a

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +18 -2
README.md CHANGED
@@ -121,6 +121,8 @@ The model was trained using the [ServiceNow/BigDocs-Sketch2Flow](https://hugging
121
  * Trained with **16× NVIDIA H100 80GB GPUs** across two nodes.
122
  * Full Sharded Data Parallel (FSDP) training, no CPU offloading.
123
 
 
 
124
  ## Evaluation
125
 
126
  ### Testing Data
@@ -156,11 +158,15 @@ All Evaluation metrics can be found in the official [StarFlow repo](https://gith
156
 
157
  Finetuning yields **large gains over base Pixtral-12B and GPT-4o**, particularly in matching workflow components and triggers.
158
 
 
 
159
  ## Model Examination
160
 
161
  * Finetuned models capture **naming conventions** and structured execution logic better.
162
  * Failure modes include **missing ELSE branches** or **generic table names**.
163
 
 
 
164
  ## Technical Specifications
165
 
166
  ### Model Architecture and Objective
@@ -173,6 +179,8 @@ Finetuning yields **large gains over base Pixtral-12B and GPT-4o**, particularly
173
  * **Hardware:** 16× NVIDIA H100 80GB (2 nodes)
174
  * **Software:** FSDP, bf16 mixed precision, PyTorch/Transformers
175
 
 
 
176
  ## Citation
177
 
178
  **BibTeX:**
@@ -189,6 +197,8 @@ Finetuning yields **large gains over base Pixtral-12B and GPT-4o**, particularly
189
  **APA:**
190
  Béchard, P., Wang, C., Abaskohi, A., Rodriguez, J., Pal, C., Vazquez, D., Gella, S., Rajeswar, S., & Taslakian, P. (2025). **StarFlow: Generating Structured Workflow Outputs from Sketch Images**. *arXiv preprint arXiv:2503.21889*.
191
 
 
 
192
  ## Glossary
193
 
194
  * **FlowSim**: Metric based on tree edit distance for workflows.
@@ -196,16 +206,22 @@ Béchard, P., Wang, C., Abaskohi, A., Rodriguez, J., Pal, C., Vazquez, D., Gella
196
  * **Trigger Match**: Correctness of predicted workflow trigger.
197
  * **Component Match**: Correctness of predicted components (order-agnostic).
198
 
 
 
199
  ## More Information
200
 
201
  * [ServiceNow Flow Designer](https://www.servicenow.com/products/platform-flow-designer.html)
202
- * Related datasets: **synthetic, manual, digital, UI-rendered** workflows.
 
 
203
 
204
  ## The StarFlow Team
205
 
206
  * Patrice Béchard, Chao Wang, Amirhossein Abaskohi, Juan Rodriguez, Christopher Pal, David Vazquez, Spandana Gella, Sai Rajeswar, Perouz Taslakian
207
 
 
 
208
  ## Model Card Contact
209
 
210
  * Patrice Bechard - [patrice.bechard@servicenow.com](mailto:patrice.bechard@servicenow.com)
211
- * ServiceNow Research – [research.servicenow.com](https://research.servicenow.com)
 
121
  * Trained with **16× NVIDIA H100 80GB GPUs** across two nodes.
122
  * Full Sharded Data Parallel (FSDP) training, no CPU offloading.
123
 
124
+ ---
125
+
126
  ## Evaluation
127
 
128
  ### Testing Data
 
158
 
159
  Finetuning yields **large gains over base Pixtral-12B and GPT-4o**, particularly in matching workflow components and triggers.
160
 
161
+ ---
162
+
163
  ## Model Examination
164
 
165
  * Finetuned models capture **naming conventions** and structured execution logic better.
166
  * Failure modes include **missing ELSE branches** or **generic table names**.
167
 
168
+ ---
169
+
170
  ## Technical Specifications
171
 
172
  ### Model Architecture and Objective
 
179
  * **Hardware:** 16× NVIDIA H100 80GB (2 nodes)
180
  * **Software:** FSDP, bf16 mixed precision, PyTorch/Transformers
181
 
182
+ ---
183
+
184
  ## Citation
185
 
186
  **BibTeX:**
 
197
  **APA:**
198
  Béchard, P., Wang, C., Abaskohi, A., Rodriguez, J., Pal, C., Vazquez, D., Gella, S., Rajeswar, S., & Taslakian, P. (2025). **StarFlow: Generating Structured Workflow Outputs from Sketch Images**. *arXiv preprint arXiv:2503.21889*.
199
 
200
+ ---
201
+
202
  ## Glossary
203
 
204
  * **FlowSim**: Metric based on tree edit distance for workflows.
 
206
  * **Trigger Match**: Correctness of predicted workflow trigger.
207
  * **Component Match**: Correctness of predicted components (order-agnostic).
208
 
209
+ ---
210
+
211
  ## More Information
212
 
213
  * [ServiceNow Flow Designer](https://www.servicenow.com/products/platform-flow-designer.html)
214
+ * [StarFlow Blog](https://www.servicenow.com/blogs/2025/starflow-ai-turns-sketches-into-workflows)
215
+
216
+ ---
217
 
218
  ## The StarFlow Team
219
 
220
  * Patrice Béchard, Chao Wang, Amirhossein Abaskohi, Juan Rodriguez, Christopher Pal, David Vazquez, Spandana Gella, Sai Rajeswar, Perouz Taslakian
221
 
222
+ ---
223
+
224
  ## Model Card Contact
225
 
226
  * Patrice Bechard - [patrice.bechard@servicenow.com](mailto:patrice.bechard@servicenow.com)
227
+ * ServiceNow Research – [research.servicenow.com](https://research.servicenow.com)