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": 75317.44081620287,
-      "required_pages_per_second_upper_estimate": 0.00652832487963992
+      "required_token_positions_per_second": 75517.52398650756,
+      "required_pages_per_second_upper_estimate": 0.006545667581735771
     },
     {
       "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": 9414.680102025359,
-      "required_pages_per_second_upper_estimate": 0.003755656654709334
+      "required_token_positions_per_second": 9439.690498313445,
+      "required_pages_per_second_upper_estimate": 0.0037656336757273994
     },
     {
       "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": 18829.360204050718,
-      "required_pages_per_second_upper_estimate": 0.00435584109778576
+      "required_token_positions_per_second": 18879.38099662689,
+      "required_pages_per_second_upper_estimate": 0.004367412527812375
     },
     {
       "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": 28244.04030607608,
-      "required_pages_per_second_upper_estimate": 0.0017199138696128897
+      "required_token_positions_per_second": 28319.071494940334,
+      "required_pages_per_second_upper_estimate": 0.0017244828753564983
     },
     {
       "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": 28244.04030607608,
-      "required_pages_per_second_upper_estimate": 0.002174304407269335
+      "required_token_positions_per_second": 28319.071494940334,
+      "required_pages_per_second_upper_estimate": 0.0021800805158877293
     },
     {
       "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": 28244.04030607608,
-      "required_pages_per_second_upper_estimate": 0.3506964538792863
+      "required_token_positions_per_second": 28319.071494940334,
+      "required_pages_per_second_upper_estimate": 0.35162809013174484
     }
   ],
   "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": 0.5146606151879496,
-    "producer_p95_seconds": 0.3736343909986317,
-    "gpu_update_seconds": 0.9788967761121679,
-    "producer_batches_per_second": 1.9430280275765976,
-    "gpu_batches_per_second": 1.0215581707926822,
-    "producer_headroom": 1.9020238720903158,
+    "producer_batch_seconds": 0.13800892956231792,
+    "producer_p95_seconds": 0.3763948949999758,
+    "gpu_update_seconds": 0.9763031957082269,
+    "producer_batches_per_second": 7.2459079508217625,
+    "gpu_batches_per_second": 1.024271972473247,
+    "producer_headroom": 7.0742030881949365,
     "sustainable": true,
-    "buffer_seconds": 60.691600118954405,
+    "buffer_seconds": 60.53079813391007,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 445308.61714717565,
-    "upload_bytes_per_second": 6137398.893169638,
+    "generation_bytes_per_second": 429686.53543145955,
+    "upload_bytes_per_second": 6244846.608886034,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.55286298572018,
+    "estimated_full_shards_per_hour": 92.2007279129776,
     "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": 0.9442589440004667,
-      "bytes": 6417253,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9173370479984442,
       "bytes": 2397470,
@@ -562,6 +556,12 @@
       "bytes": 807847,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.4682032510027057,
+      "bytes": 16949786,
+      "pages": 39,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -328,3 +328,4 @@
 {"step": 32800, "loss": 0.6038375701755285, "lr": 0.0003, "grad_norm": 0.14351599564844275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4800038111, "tokens_per_second": 149519.52894199218, "data_wait_s": 1.7172998923342675e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 32900, "loss": 0.6561686396598816, "lr": 0.0003, "grad_norm": 0.14103283728292706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4814788163, "tokens_per_second": 150550.10515941886, "data_wait_s": 9.75099828792736e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 33000, "loss": 0.6700921505689621, "lr": 0.0003, "grad_norm": 0.13229987457814377, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4829468597, "tokens_per_second": 149981.7103048597, "data_wait_s": 8.353999874088913e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 33100, "loss": 0.6747942809015512, "lr": 0.0003, "grad_norm": 0.14870925417983244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4843985769, "tokens_per_second": 148094.7395123819, "data_wait_s": 8.624003385193646e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:77db8add18adca02fc02d2aeb02379c44b8c617ffaa197c14e2c022719042794
-size 698029
+oid sha256:6458951b64cd8f37a9693ea7ea0607f349d4344cee81a899145e619da37761be
+size 700147