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": 75239.17562417206,
-      "required_pages_per_second_upper_estimate": 0.006521541051156005
+      "required_token_positions_per_second": 75728.01195696587,
+      "required_pages_per_second_upper_estimate": 0.006563912145538214
     },
     {
       "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": 9404.896953021507,
-      "required_pages_per_second_upper_estimate": 0.0037517540103005854
+      "required_token_positions_per_second": 9466.001494620734,
+      "required_pages_per_second_upper_estimate": 0.0037761295255388278
     },
     {
       "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": 18809.793906043014,
-      "required_pages_per_second_upper_estimate": 0.004351314779096761
+      "required_token_positions_per_second": 18932.002989241468,
+      "required_pages_per_second_upper_estimate": 0.004379585699688319
     },
     {
       "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": 28214.690859064525,
-      "required_pages_per_second_upper_estimate": 0.0017181266422745338
+      "required_token_positions_per_second": 28398.0044838622,
+      "required_pages_per_second_upper_estimate": 0.0017292894802524528
     },
     {
       "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": 28214.690859064525,
-      "required_pages_per_second_upper_estimate": 0.0021720450055939164
+      "required_token_positions_per_second": 28398.0044838622,
+      "required_pages_per_second_upper_estimate": 0.0021861569958754328
     },
     {
       "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": 28214.690859064525,
-      "required_pages_per_second_upper_estimate": 0.35033203197368323
+      "required_token_positions_per_second": 28398.0044838622,
+      "required_pages_per_second_upper_estimate": 0.3526081736824342
     }
   ],
   "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": 0.28985642799990785,
-    "producer_p95_seconds": 0.23285877899979823,
-    "gpu_update_seconds": 0.9799150427734542,
-    "producer_batches_per_second": 3.44998386580655,
-    "gpu_batches_per_second": 1.0204966311872297,
-    "producer_headroom": 3.380691087429552,
+    "producer_batch_seconds": 0.06483750674210853,
+    "producer_p95_seconds": 0.21952647500074818,
+    "gpu_update_seconds": 0.9735895356911994,
+    "producer_batches_per_second": 15.423171714136148,
+    "gpu_batches_per_second": 1.02712689828784,
+    "producer_headroom": 15.015838588051452,
     "sustainable": true,
-    "buffer_seconds": 60.75473265195416,
+    "buffer_seconds": 60.36255121285436,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 376790.25279944093,
-    "upload_bytes_per_second": 5897650.912680773,
+    "generation_bytes_per_second": 376835.85868988116,
+    "upload_bytes_per_second": 5963202.5255215475,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.85041702258512,
+    "estimated_full_shards_per_hour": 80.8602029850228,
     "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": 0.7197847669995099,
-      "bytes": 852888,
-      "pages": 25,
-      "shards": 1
-    },
     {
       "seconds": 0.9669863799999803,
       "bytes": 6140852,
@@ -562,6 +556,12 @@
       "bytes": 2312576,
       "pages": 26,
       "shards": 2
+    },
+    {
+      "seconds": 1.5429994639998768,
+      "bytes": 10370731,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -128,3 +128,4 @@
 {"step": 127800, "loss": 0.5831902325153351, "lr": 0.0003, "grad_norm": 0.13881338642355742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18703401033, "tokens_per_second": 150540.16901954522, "data_wait_s": 1.344199881714303e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 127900, "loss": 0.6205321792513132, "lr": 0.0003, "grad_norm": 0.14496050650069636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18717891551, "tokens_per_second": 147612.9805286436, "data_wait_s": 1.391300065733958e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 128000, "loss": 0.6035799086093903, "lr": 0.0003, "grad_norm": 0.13391310087245253, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18732618299, "tokens_per_second": 150480.36727232047, "data_wait_s": 1.367599907098338e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 128100, "loss": 0.6263636723160744, "lr": 0.0003, "grad_norm": 0.13098580312341396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18747193942, "tokens_per_second": 148760.9482322472, "data_wait_s": 1.6726000467315316e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a150057152ee63d678cc9c09c53a0fd1d932b2b4b56502f3f7d6e0bdeccbfa56
-size 277071
+oid sha256:13c80b258d149581f19eb3b36f77b7bc2d2934425ad83c408fe0df3909b243eb
+size 279189