Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75399.42172020269,
-      "required_pages_per_second_upper_estimate": 0.006535430776619923
+      "required_token_positions_per_second": 75331.0605890886,
+      "required_pages_per_second_upper_estimate": 0.006529505407034657
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9424.927715025337,
-      "required_pages_per_second_upper_estimate": 0.0037597445807504935
+      "required_token_positions_per_second": 9416.382573636076,
+      "required_pages_per_second_upper_estimate": 0.003756335796089068
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18849.855430050673,
-      "required_pages_per_second_upper_estimate": 0.004360582307617191
+      "required_token_positions_per_second": 18832.76514727215,
+      "required_pages_per_second_upper_estimate": 0.004356628771475085
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28274.78314507601,
-      "required_pages_per_second_upper_estimate": 0.0017217859472127746
+      "required_token_positions_per_second": 28249.14772090823,
+      "required_pages_per_second_upper_estimate": 0.001720224884379633
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28274.78314507601,
-      "required_pages_per_second_upper_estimate": 0.002176671076117175
+      "required_token_positions_per_second": 28249.14772090823,
+      "required_pages_per_second_upper_estimate": 0.002174697590201343
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28274.78314507601,
-      "required_pages_per_second_upper_estimate": 0.351078177050002
+      "required_token_positions_per_second": 28249.14772090823,
+      "required_pages_per_second_upper_estimate": 0.35075987087808375
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.24759402542196085,
-    "producer_p95_seconds": 0.3587101380035165,
-    "gpu_update_seconds": 0.9778324331663296,
-    "producer_batches_per_second": 4.038869671009852,
-    "gpu_batches_per_second": 1.022670107967159,
-    "producer_headroom": 3.949337757645257,
+    "producer_batch_seconds": 0.09008804580486185,
+    "producer_p95_seconds": 0.3530405320052523,
+    "gpu_update_seconds": 0.9787197926518931,
+    "producer_batches_per_second": 11.10025188209857,
+    "gpu_batches_per_second": 1.0217429007851644,
+    "producer_headroom": 10.8640362204313,
     "sustainable": true,
-    "buffer_seconds": 60.625610856312434,
+    "buffer_seconds": 60.68062714441737,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 416267.331035606,
-    "upload_bytes_per_second": 6126148.728639764,
+    "generation_bytes_per_second": 416342.30290317297,
+    "upload_bytes_per_second": 6010721.471126111,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.32127903271804,
+    "estimated_full_shards_per_hour": 89.33736625024216,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.1623649149987614,
-      "bytes": 11298342,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.995095141002821,
       "bytes": 6689673,
@@ -562,6 +556,12 @@
       "bytes": 17347426,
       "pages": 42,
       "shards": 5
+    },
+    {
+      "seconds": 0.7935998479952104,
+      "bytes": 915809,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -298,3 +298,4 @@
 {"step": 114800, "loss": 0.5815488835796714, "lr": 0.0003, "grad_norm": 0.1341264427182585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16799325296, "tokens_per_second": 149237.73573769786, "data_wait_s": 1.0627998562995344e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 114900, "loss": 0.6156587619334459, "lr": 0.0003, "grad_norm": 0.13408592870255762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16813972431, "tokens_per_second": 149642.518388121, "data_wait_s": 1.3570999726653099e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 115000, "loss": 0.6215994767844677, "lr": 0.0003, "grad_norm": 0.14603409049989743, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16828383176, "tokens_per_second": 146871.65805112463, "data_wait_s": 1.9731000065803528e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 115100, "loss": 0.6090226266533136, "lr": 0.0003, "grad_norm": 0.12995592805320222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16843057721, "tokens_per_second": 133816.77263533513, "data_wait_s": 1.7527003365103155e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f4db7491a8f8dcfa529312e197a73bd785c376701bd9f4dd2b42df5e1cba763c
-size 640263
+oid sha256:259bfddd9c44668ace0115211018936a13600e6fcd8ba37d7e58fb4625780222
+size 642381