Update README.md

README.md

@@ -9,30 +9,26 @@ This is the final assignment it synthesizes Assignments 2 and 3 into a data labe
 ## Pipeline Architecture
 
 patents_50k_green.parquet
-        │
-        ▼
+
 [Part A & B] Baseline PatentSBERTa + Uncertainty Sampling
-        │  → Top 100 high-risk claims (u ≈ 1.0)
-        ▼
+- Top 100 high-risk claims (u ≈ 1.0)
+
 [Part C – Step 1] QLoRA Fine-tuning on Colab (Qwen3-8B, 4-bit, 3 epochs)
-        │  → qlora_green_patent_adapter  (LoRA weights)
-        │  → Qwen3-8B.Q4_K_M.gguf      (served via LM Studio)
-        ▼
+- qlora_green_patent_adapter  (LoRA weights)
+- Qwen3-8B.Q4_K_M.gguf      (served via LM Studio)
+
 [Part C – Step 2] Multi-Agent System (CrewAI)
-        │  Agent 1 – Advocate  (Qwen3-4B, argues for green: Advocator)
-        │  Agent 2 – Skeptic   (Qwen3-4B, argues against green: Skeptic)
-        │  Agent 3 – Judge     (QLoRA Qwen3-8B, final verdict: Judge)
-        ▼
+- Agent 1 – Advocate  (Qwen3-4B, argues for green: Advocator)
+- Agent 2 – Skeptic   (Qwen3-4B, argues against green: Skeptic)
+-  Agent 3 – Judge     (QLoRA Qwen3-8B, final verdict: Judge)
+
 [Part D] Exception-Based HITL (only deadlocks / low-confidence)
-        │  → 26 claims reviewed with deadlock, 3 human overrides
-        │  → hitl_green_100_final.csv  (gold labels)
-        ▼
-[Part D] Final PatentSBERTa Fine-tuning on gold dataset
-           → patentsberta_finetuned_final/
-        
+- 26 claims reviewed with deadlock, 3 human overrides
+- hitl_green_100_final.csv  (gold labels)
 
+[Part D] Final PatentSBERTa Fine-tuning on gold dataset
+- patentsberta_finetuned_final/
 
----
 
 ## Part C – Step 1: QLoRA Domain Adaptation
 
@@ -55,7 +51,6 @@ The generative LLM fine-tuning was performed on Google Colab (T4, 15 GB VRAM) us
 
 The fine-tuned adapter was exported to GGUF Q4_K_M format (4.682 GB) and served locally via LM Studio for use in the MAS.
 
----
 
 ## Part C – Step 2: Multi-Agent System
 
@@ -69,7 +64,6 @@ Three agents collaborate to label each of the 100 high-risk patent claims using
 
 Each claim produces a structured JSON output: `classification` (0/1), `confidence` (Low/Medium/High), and `rationale`.
 
----
 
 ## Part D: Targeted HITL & Final Fine-tuning
 
@@ -85,7 +79,6 @@ Each claim produces a structured JSON output: `classification` (0/1), `confidenc
 
 The gold-labelled dataset (`hitl_green_100_final.csv`) was used to fine-tune PatentSBERTa for 3 epochs using CosineSimilarityLoss on an AMD Radeon RX 9070 XT via DirectML (fell back to CPU, completed in ~31 minutes).
 
----
 
 ## Performance Results
 
@@ -98,7 +91,6 @@ The gold-labelled dataset (`hitl_green_100_final.csv`) was used to fine-tune Pat
 
 The Final Model achieves the highest F1 score across all iterations, demonstrating that QLoRA domain adaptation combined with structured agent debate and targeted human review produces measurable improvements.
 
----
 
 ## Key Findings
 
@@ -117,7 +109,6 @@ The Final Model achieves the highest F1 score across all iterations, demonstrati
 - torchao 0.16.0 conflicts with transformers lazy loader in certain environments
 - LM Studio local inference adds latency (~3–5s per agent call)
 
----
 
 ## Repository Contents
 
@@ -130,7 +121,6 @@ The Final Model achieves the highest F1 score across all iterations, demonstrati
 | `qlora_outputs.zip` | QLoRA adapter weights (`qlora_green_patent_adapter/`) |
 | `Part C Step 1.ipynb` | Colab notebook for QLoRA fine-tuning |
 
----
 
 ## Related Repositories