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": 75355.77308669794,
-      "required_pages_per_second_upper_estimate": 0.006531647423694181
+      "required_token_positions_per_second": 75493.4830562282,
+      "required_pages_per_second_upper_estimate": 0.006543583774830337
     },
     {
       "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": 9419.471635837243,
-      "required_pages_per_second_upper_estimate": 0.003757568069186709
+      "required_token_positions_per_second": 9436.685382028525,
+      "required_pages_per_second_upper_estimate": 0.0037644348899108526
     },
     {
       "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": 18838.943271674485,
-      "required_pages_per_second_upper_estimate": 0.004358057971824289
+      "required_token_positions_per_second": 18873.37076405705,
+      "required_pages_per_second_upper_estimate": 0.004366022166283823
     },
     {
       "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": 28258.414907511728,
-      "required_pages_per_second_upper_estimate": 0.0017207892074226155
+      "required_token_positions_per_second": 28310.05614608557,
+      "required_pages_per_second_upper_estimate": 0.001723933887911123
     },
     {
       "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": 28258.414907511728,
-      "required_pages_per_second_upper_estimate": 0.002175411003879295
+      "required_token_positions_per_second": 28310.05614608557,
+      "required_pages_per_second_upper_estimate": 0.002179386489376807
     },
     {
       "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": 28258.414907511728,
-      "required_pages_per_second_upper_estimate": 0.35087493832042077
+      "required_token_positions_per_second": 28310.05614608557,
+      "required_pages_per_second_upper_estimate": 0.3515161496714004
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.21612502935144562,
-    "producer_p95_seconds": 0.34198722200017073,
-    "gpu_update_seconds": 0.9783988270570171,
-    "producer_batches_per_second": 4.626951366997287,
-    "gpu_batches_per_second": 1.0220780854858118,
-    "producer_headroom": 4.527003790320007,
+    "producer_batch_seconds": 0.08282781931251293,
+    "producer_p95_seconds": 0.3504311319993576,
+    "gpu_update_seconds": 0.9766140998566294,
+    "producer_batches_per_second": 12.073238294816829,
+    "gpu_batches_per_second": 1.0239458964874701,
+    "producer_headroom": 11.790894749647125,
     "sustainable": true,
-    "buffer_seconds": 60.66072727753506,
+    "buffer_seconds": 60.55007419111102,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 465182.8805130567,
+    "generation_bytes_per_second": 465752.25341577077,
     "upload_bytes_per_second": 4179249.6480450425,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.81741725486542,
+    "estimated_full_shards_per_hour": 99.93959142546504,
     "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": [

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

@@ -237,3 +237,4 @@
 {"step": 108700, "loss": 0.5989245846867561, "lr": 0.0003, "grad_norm": 0.1259789825213902, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15906098309, "tokens_per_second": 147376.48087369718, "data_wait_s": 1.2828997569158673e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 108800, "loss": 0.6310719847679138, "lr": 0.0003, "grad_norm": 0.1289245080110836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15920844965, "tokens_per_second": 149894.3880822485, "data_wait_s": 1.9346000044606626e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 108900, "loss": 0.598317040130496, "lr": 0.0003, "grad_norm": 0.13598713407394408, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15935566462, "tokens_per_second": 149171.47226963693, "data_wait_s": 1.5513000107603148e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 109000, "loss": 0.6347799599170685, "lr": 0.0003, "grad_norm": 0.13985157473246534, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15950330622, "tokens_per_second": 149699.75460610056, "data_wait_s": 1.5652000001864508e-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 sha256:2d8df95e78706515868a040c4edabf3790b2e8157b7a6570d169f8637e520996
-size 509499
+oid sha256:6fabc53c4267b5894f1fe71899e3bda1c17829ecdfd769745f3dd0e3647ef0aa
+size 511617