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": 75908.44351840722,
-      "required_pages_per_second_upper_estimate": 0.006579551495984339
+      "required_token_positions_per_second": 75954.54383687535,
+      "required_pages_per_second_upper_estimate": 0.0065835473547489675
     },
     {
       "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": 9488.555439800903,
-      "required_pages_per_second_upper_estimate": 0.003785126631482728
+      "required_token_positions_per_second": 9494.31797960942,
+      "required_pages_per_second_upper_estimate": 0.0037874253947699933
     },
     {
       "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": 18977.110879601805,
-      "required_pages_per_second_upper_estimate": 0.004390020616251426
+      "required_token_positions_per_second": 18988.63595921884,
+      "required_pages_per_second_upper_estimate": 0.004392686740586353
     },
     {
       "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": 28465.66631940271,
-      "required_pages_per_second_upper_estimate": 0.001733409731042654
+      "required_token_positions_per_second": 28482.953938828257,
+      "required_pages_per_second_upper_estimate": 0.0017344624563645405
     },
     {
       "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": 28465.66631940271,
-      "required_pages_per_second_upper_estimate": 0.0021913657912752895
+      "required_token_positions_per_second": 28482.953938828257,
+      "required_pages_per_second_upper_estimate": 0.0021926966400738616
     },
     {
       "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": 28465.66631940271,
-      "required_pages_per_second_upper_estimate": 0.35344830723025084
+      "required_token_positions_per_second": 28482.953938828257,
+      "required_pages_per_second_upper_estimate": 0.3536629616055758
     }
   ],
   "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": 0.5297710319763382,
-    "producer_p95_seconds": 0.31004507200123044,
-    "gpu_update_seconds": 0.9712753493901046,
-    "producer_batches_per_second": 1.8876079280315654,
-    "gpu_batches_per_second": 1.0295741579645077,
-    "producer_headroom": 1.8333870498103901,
+    "producer_batch_seconds": 0.1653632896328645,
+    "producer_p95_seconds": 0.35368149500209256,
+    "gpu_update_seconds": 0.9706858375496638,
+    "producer_batches_per_second": 6.0472914043992185,
+    "gpu_batches_per_second": 1.0301994335513691,
+    "producer_headroom": 5.870020121786138,
     "sustainable": true,
-    "buffer_seconds": 60.21907166218649,
+    "buffer_seconds": 60.182521928079154,
     "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": 299961.92679126974,
-    "upload_bytes_per_second": 3965305.3179462114,
+    "generation_bytes_per_second": 287573.6403975973,
+    "upload_bytes_per_second": 4002366.2795159263,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.3648467331273,
+    "estimated_full_shards_per_hour": 61.706608857592954,
     "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": 1.1714967369989608,
-      "bytes": 6024956,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.1858630629976687,
       "bytes": 13358845,
@@ -562,6 +556,12 @@
       "bytes": 661491,
       "pages": 19,
       "shards": 1
+    },
+    {
+      "seconds": 1.42470605500057,
+      "bytes": 11031230,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -247,3 +247,4 @@
 {"step": 139700, "loss": 0.6310807839035988, "lr": 0.0003, "grad_norm": 0.15356001626441015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20445353412, "tokens_per_second": 149751.25880204746, "data_wait_s": 1.5083998732734472e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 139800, "loss": 0.5826917011290789, "lr": 0.0003, "grad_norm": 0.13940416124361288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20459865726, "tokens_per_second": 126068.19789574093, "data_wait_s": 1.4616998669225723e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 139900, "loss": 0.5581844821572304, "lr": 0.0003, "grad_norm": 0.15211728827907198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20474365976, "tokens_per_second": 147882.59859125703, "data_wait_s": 1.2952001270605251e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 140000, "loss": 0.6059972457587719, "lr": 0.0003, "grad_norm": 0.12777334099873983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20489021311, "tokens_per_second": 149128.76249265246, "data_wait_s": 1.251399953616783e-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 sha256:9a3ef07b033c6f341ef57ad01e35af7a4c7800821f44791ff3e670251f967c7a
-size 530679
+oid sha256:d149754d7fe982e05cd4a6754bf7ed3c090591faae7676f16c9cdcb6b0a6218d
+size 532797