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": 75231.73542584629,
-      "required_pages_per_second_upper_estimate": 0.0065208961536221376
+      "required_token_positions_per_second": 75245.04306715509,
+      "required_pages_per_second_upper_estimate": 0.0065220496262428755
     },
     {
       "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": 9403.966928230786,
-      "required_pages_per_second_upper_estimate": 0.003751383009506457
+      "required_token_positions_per_second": 9405.630383394386,
+      "required_pages_per_second_upper_estimate": 0.003752046586642088
     },
     {
       "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": 18807.933856461572,
-      "required_pages_per_second_upper_estimate": 0.004350884489361849
+      "required_token_positions_per_second": 18811.26076678877,
+      "required_pages_per_second_upper_estimate": 0.004351654111514425
     },
     {
       "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": 28211.90078469236,
-      "required_pages_per_second_upper_estimate": 0.0017179567413836579
+      "required_token_positions_per_second": 28216.891150183157,
+      "required_pages_per_second_upper_estimate": 0.0017182606284596232
     },
     {
       "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": 28211.90078469236,
-      "required_pages_per_second_upper_estimate": 0.0021718302179453267
+      "required_token_positions_per_second": 28216.891150183157,
+      "required_pages_per_second_upper_estimate": 0.0021722143900950244
     },
     {
       "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": 28211.90078469236,
-      "required_pages_per_second_upper_estimate": 0.35029738858775916
+      "required_token_positions_per_second": 28216.891150183157,
+      "required_pages_per_second_upper_estimate": 0.350359352225476
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.13236735547641842,
-    "producer_p95_seconds": 0.30508852700222633,
-    "gpu_update_seconds": 0.9800119535016113,
-    "producer_batches_per_second": 7.554732784384686,
-    "gpu_batches_per_second": 1.0203957170389308,
-    "producer_headroom": 7.403728434207503,
+    "producer_batch_seconds": 0.08041407803887068,
+    "producer_p95_seconds": 0.2778208350027853,
+    "gpu_update_seconds": 0.9798386311534018,
+    "producer_batches_per_second": 12.435633466028404,
+    "gpu_batches_per_second": 1.0205762134759533,
+    "producer_headroom": 12.184914072878705,
     "sustainable": true,
-    "buffer_seconds": 60.7607411170999,
+    "buffer_seconds": 60.74999513151091,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 325017.95320412173,
-    "upload_bytes_per_second": 5400581.828916083,
+    "generation_bytes_per_second": 325222.3081529987,
+    "upload_bytes_per_second": 5516253.685440331,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.74128672688235,
+    "estimated_full_shards_per_hour": 69.7851365417716,
     "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": 2.3653303009996307,
-      "bytes": 1160564,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.3003170710017002,
       "bytes": 11354768,
@@ -562,6 +556,12 @@
       "bytes": 5691792,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.035739069000556,
+      "bytes": 2751840,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -224,3 +224,4 @@
 {"step": 337400, "loss": 0.5923497267067432, "lr": 0.0003, "grad_norm": 0.15710091453944813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49386590651, "tokens_per_second": 149696.4040228238, "data_wait_s": 1.3985998521093279e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 337500, "loss": 0.6300313007086515, "lr": 0.0003, "grad_norm": 0.1427950139887913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49401225614, "tokens_per_second": 148533.7911642823, "data_wait_s": 1.0522999218665063e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 337600, "loss": 0.6027877815067768, "lr": 0.0003, "grad_norm": 0.14074400411154614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49415928235, "tokens_per_second": 148568.92948110442, "data_wait_s": 1.4468998415395617e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 337700, "loss": 0.5835673455148935, "lr": 0.0003, "grad_norm": 0.1706562095305728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49430605402, "tokens_per_second": 148915.6859895565, "data_wait_s": 1.136900027631782e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:870bdcb25542580f053fb120cc037030acf4511cf6fb166da2111a403d9fbfa6
-size 481965
+oid sha256:36060898808f7170492c9bb2d0ead309af344355f6a449f1b587242a082741cf
+size 484083