Add flops information (round 4)

Author: DBlankvoort

33x-coder.yaml

@@ -14,7 +14,7 @@
 # - Contributor type: Non-academic (Research community)
 
 # Training compute:
-# - Base model training compute: ~4.08e+23 FLOP +- 0.5 OoM (Epoch AI)
+# - Base model training compute: ~4.08e+23 FLOP +- 0.5 OoM (param count & dataset size) [Epoch AI]
 # - End model training compute: unknown (likely negligible)
 
 system:

Arabic StableLM.yaml

@@ -14,7 +14,7 @@
 # - Contributor type: Non-academic (Company)
 
 # Training compute:
-# - Base model training compute: ~1.92e+22 FLOP +- 0.5 OoM (Epoch AI)
+# - Base model training compute: ~1.92e+22 FLOP +- 0.5 OoM (param count & dataset site) [Epoch AI]
 # - End model training compute: unknown (likely negligible)
 
 system:

BELLE.yaml

@@ -14,7 +14,7 @@
 # - Contributor type: Non-academic (Company)
 
 # Training compute:
-# - Base model training compute: ~1.6e+23 FLOP +- 0.5 OoM (Epoch AI)
+# - Base model training compute: ~1.6e+23 FLOP +- 0.5 OoM (param count & dataset size) [Epoch AI]
 # - End model training compute: unknown (likely negligible)
 
 system:

CT-LLM.yaml

@@ -13,6 +13,10 @@
 # - National origin: China
 # - Contributor type: Academic (Research community)
 
+# Training compute:
+# - Base model training compute: >= 1.81e+22 FLOP +- 0.5 OoM (param count & dataset size)
+# - End model training compute: unknown (likely negligible)
+
 system:
     name: CT-LLM
     link: https://huggingface.co/m-a-p/CT-LLM-SFT-DPO

ChatMusician.yaml

@@ -14,15 +14,15 @@
 # - Contributor type: Academic (Research community)
 
 # Training compute:
-# - Base model training compute: ~8.4e+22 FLOP +- 0.5 OoM (Epoch AI)
+# - Base model training compute: ~8.4e+22 FLOP +- 0.5 OoM (param count & dataset size) [Epoch AI]
 # - End model training compute: unknown (likely negligible)
 
 system:
     name: ChatMusician
     link: https://huggingface.co/m-a-p/ChatMusician
     type: audio
     performanceclass: limited
-    basemodelname: ChatMusician-Base
+    basemodelname: Llama-2-7B
     endmodelname: ChatMusician
     endmodellicense: MIT License
     releasedate: 2024-04
@@ -35,19 +35,19 @@ org:
 
 # availability:
 datasources_basemodel:
-    class: partial
-    link: https://huggingface.co/datasets/m-a-p/MusicPile
-    notes:
+    class: closed
+    link: 
+    notes: No dataset information found.
 
 datasources_endmodel:
     class: partial
     link: https://huggingface.co/datasets/m-a-p/MusicPile
     notes: 
 
 weights_basemodel:
-    class: open
-    link: https://huggingface.co/m-a-p/ChatMusician
-    notes:
+    class: partial
+    link: https://huggingface.co/meta-llama/Llama-2-7b
+    notes: Gated model on HuggingFace.
 
 weights_endmodel:
     class: open

Claire.yaml

@@ -14,7 +14,7 @@
 # - Contributor type: Academic (Research community)
 
 # Training compute:
-# - Base model training compute: ~6.3e+22 FLOP +- 0.5 OoM (Epoch AI)
+# - Base model training compute: ~6.3e+22 FLOP +- 0.5 OoM (both param count & dataset site, and hardware figures) [Epoch AI]
 # - End model training compute: unknown (likely negligible)
 
 system:

CodeGeeX.yaml

@@ -14,7 +14,7 @@
 # - Contributor type: Non-academic (Chinese AI Tiger)
 
 # Training compute:
-# - Base model training compute: ~6.63e+22 FLOP +- 0.5 OoM (Epoch AI)
+# - Base model training compute: ~8.1e+23 FLOP +- 1.0 OoM (param count & dataset size) [Epoch AI]
 # - End model training compute: unknown (likely negligible)
 
 system:

CodeGemma.yaml

@@ -14,7 +14,7 @@
 # - Contributor type: Non-academic (American Big Tech)
 
 # Training compute:
-# - Base model training compute: ~3.07e+23 FLOP +- 0.5 OoM (Epoch AI)
+# - Base model training compute: ~3.07e+23 FLOP +- 0.5 OoM (param count & dataset size) [Epoch AI]
 # - End model training compute: unknown (likely negligible)
 
 system:

CodeLlama.yaml

@@ -14,8 +14,8 @@
 # - Contributor type: Non-academic (American Big Tech)
 
 # Training compute:
-# - Base model training compute: ~8.1e+23 FLOP +- 0.5 OoM (Epoch AI)
-# - End model training compute: ~4.2e+23 FLOP +- 0.3 OoM (param count & dataset size)
+# - Base model training compute: ~8.1e+23 FLOP +- 0.5 OoM (geometric mean of param count & dataset site, and hardware figures) [Epoch AI]
+# - End model training compute: ~4.2e+23 FLOP +- 0.5 OoM (param count & dataset size)
 
 system:
     name: CodeLlama

CodeUp.yaml

@@ -14,7 +14,7 @@
 # - Contributor type: Academic (University)
 
 # Training compute:
-# - Base model training compute: ~7.2e+23 FLOP +- 0.5 OoM (Epoch AI)
+# - Base model training compute: ~7.2e+23 FLOP +- 0.5 OoM (hardware figures) [Epoch AI]
 # - End model training compute: unknown (likely negligible)
 
 system: