Add Bielik v0.1 LLM

configs/model/Bielik-7B-Instruct-v0.1-fine-tuned.yaml

@@ -0,0 +1,11 @@
+name: speakleash/Bielik-7B-Instruct-v0.1
+tokenizer_name: ${.name}
+
+adapter_path: data/experiments/fine-tune/Bielik-7B-Instruct-v0.1/pl-court-instruct/checkpoint-1500
+
+max_seq_length: 4096
+batch_size: 1
+padding: longest
+use_4bit: true
+
+use_unsloth: true

configs/model/Bielik-7B-Instruct-v0.1.yaml

@@ -0,0 +1,11 @@
+name: speakleash/Bielik-7B-Instruct-v0.1
+tokenizer_name: ${.name}
+
+adapter_path: null
+
+max_seq_length: 4096
+batch_size: 1
+padding: longest
+use_4bit: true
+
+use_unsloth: true

configs/model/Unsloth-Mistral-Nemo-Instruct-2407-fine-tuned.yaml

@@ -1,7 +1,7 @@
 name: unsloth/Mistral-Nemo-Instruct-2407-bnb-4bit
 tokenizer_name: ${.name}
 
-adapter_path: data/experiments/fine-tune/Unsloth-Mistral-Nemo-Instruct-2407/pl-court-instruct/checkpoint-500
+adapter_path: data/experiments/fine-tune/Unsloth-Mistral-Nemo-Instruct-2407/pl-court-instruct/checkpoint-1500
 
 max_seq_length: 7_900 # can handle larger inputs, but set is equal to llama-3 for better comparison
 padding: longest

data/experiments/fine-tune/Bielik-7B-Instruct-v0.1/.gitignore

@@ -0,0 +1 @@
+/pl-court-instruct

data/experiments/predict/pl-court-instruct/Bielik-7B-Instruct-v0.1-fine-tuned/.gitignore

@@ -0,0 +1,6 @@
+/outputs_42.json
+/outputs_7312.json
+/outputs_997.json
+/metrics_42.json
+/metrics_7312.json
+/metrics_997.json

data/experiments/predict/pl-court-instruct/Bielik-7B-Instruct-v0.1/.gitignore

@@ -0,0 +1,6 @@
+/outputs_997.json
+/outputs_42.json
+/outputs_7312.json
+/metrics_42.json
+/metrics_7312.json
+/metrics_997.json

data/experiments/predict/pl-court-instruct/metrics_ngram_summary.md

@@ -1,10 +1,12 @@
 | llm                                           | full_text_chrf   | court_name      | date            | department_name   | judges          | legal_bases     | recorder        | signature       |
 |:----------------------------------------------|:-----------------|:----------------|:----------------|:------------------|:----------------|:----------------|:----------------|:----------------|
+| Bielik-7B-Instruct-v0.1                       | 0.354 (± 0.001)  | 0.000 (± 0.000) | 0.001 (± 0.000) | 0.001 (± 0.000)   | 0.001 (± 0.000) | 0.001 (± 0.000) | 0.000 (± 0.000) | 0.000 (± 0.000) |
+| Bielik-7B-Instruct-v0.1-fine-tuned            | 0.717 (± 0.000)  | 0.890 (± 0.007) | 0.863 (± 0.007) | 0.886 (± 0.007)   | 0.879 (± 0.007) | 0.465 (± 0.004) | 0.639 (± 0.001) | 0.459 (± 0.002) |
 | Unsloth-Llama-3-8B-Instruct                   | 0.579 (± 0.001)  | 0.865 (± 0.000) | 0.948 (± 0.001) | 0.882 (± 0.026)   | 0.902 (± 0.011) | 0.312 (± 0.042) | 0.741 (± 0.002) | 0.665 (± 0.022) |
 | Unsloth-Llama-3-8B-Instruct-fine-tuned        | 0.747 (± 0.000)  | 0.916 (± 0.001) | 0.920 (± 0.002) | 0.902 (± 0.000)   | 0.906 (± 0.001) | 0.442 (± 0.001) | 0.812 (± 0.003) | 0.805 (± 0.004) |
 | Unsloth-Mistral-7B-Instruct-v0.3              | 0.574 (± 0.001)  | 0.397 (± 0.005) | 0.470 (± 0.004) | 0.404 (± 0.005)   | 0.424 (± 0.003) | 0.159 (± 0.002) | 0.436 (± 0.003) | 0.159 (± 0.001) |
 | Unsloth-Mistral-7B-Instruct-v0.3-fine-tuned   | 0.634 (± 0.001)  | 0.547 (± 0.003) | 0.549 (± 0.003) | 0.543 (± 0.003)   | 0.544 (± 0.003) | 0.366 (± 0.002) | 0.534 (± 0.002) | 0.533 (± 0.001) |
 | Unsloth-Mistral-Nemo-Instruct-2407            | 0.520 (± 0.001)  | 0.732 (± 0.006) | 0.759 (± 0.005) | 0.687 (± 0.006)   | 0.619 (± 0.006) | 0.267 (± 0.002) | 0.690 (± 0.008) | 0.600 (± 0.004) |
-| Unsloth-Mistral-Nemo-Instruct-2407-fine-tuned | 0.610 (± 0.000)  | 0.782 (± 0.002) | 0.742 (± 0.000) | 0.717 (± 0.002)   | 0.717 (± 0.001) | 0.368 (± 0.000) | 0.696 (± 0.003) | 0.650 (± 0.003) |
+| Unsloth-Mistral-Nemo-Instruct-2407-fine-tuned | 0.708 (± 0.001)  | 0.900 (± 0.001) | 0.843 (± 0.000) | 0.818 (± 0.001)   | 0.826 (± 0.001) | 0.503 (± 0.002) | 0.693 (± 0.007) | 0.642 (± 0.007) |
 | open_ai_gpt-4o                                | 0.651 (± nan)    | 0.955 (± nan)   | 0.986 (± nan)   | 0.971 (± nan)     | 0.917 (± nan)   | 0.502 (± nan)   | 0.834 (± nan)   | 0.990 (± nan)   |
 | open_ai_gpt-4o-mini                           | 0.646 (± nan)    | 0.953 (± nan)   | 0.986 (± nan)   | 0.976 (± nan)     | 0.927 (± nan)   | 0.534 (± nan)   | 0.969 (± nan)   | 0.988 (± nan)   |