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": 75982.558902417,
-      "required_pages_per_second_upper_estimate": 0.0065859756296265835
+      "required_token_positions_per_second": 75011.02245252192,
+      "required_pages_per_second_upper_estimate": 0.006501765312486299
     },
     {
       "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": 9497.819862802125,
-      "required_pages_per_second_upper_estimate": 0.0037888223483333833
+      "required_token_positions_per_second": 9376.37780656524,
+      "required_pages_per_second_upper_estimate": 0.0037403772963799427
     },
     {
       "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": 18995.63972560425,
-      "required_pages_per_second_upper_estimate": 0.00439430693867754
+      "required_token_positions_per_second": 18752.75561313048,
+      "required_pages_per_second_upper_estimate": 0.004338119973871116
     },
     {
       "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": 28493.45958840638,
-      "required_pages_per_second_upper_estimate": 0.0017351021953049646
+      "required_token_positions_per_second": 28129.13341969572,
+      "required_pages_per_second_upper_estimate": 0.0017129166431021664
     },
     {
       "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": 28493.45958840638,
-      "required_pages_per_second_upper_estimate": 0.0021935053940598847
+      "required_token_positions_per_second": 28129.13341969572,
+      "required_pages_per_second_upper_estimate": 0.0021654585570731553
     },
     {
       "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": 28493.45958840638,
-      "required_pages_per_second_upper_estimate": 0.3537934066131887
+      "required_token_positions_per_second": 28129.13341969572,
+      "required_pages_per_second_upper_estimate": 0.34926969491905235
     }
   ],
   "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": 0.08790515790622067,
-    "producer_p95_seconds": 0.204653488999611,
-    "gpu_update_seconds": 0.9703279418989759,
-    "producer_batches_per_second": 11.375896748480038,
-    "gpu_batches_per_second": 1.0305794121964111,
-    "producer_headroom": 11.038350479207887,
+    "producer_batch_seconds": 0.25700118089062585,
+    "producer_p95_seconds": 0.2640638319971913,
+    "gpu_update_seconds": 0.9828955477398804,
+    "producer_batches_per_second": 3.891032704731339,
+    "gpu_batches_per_second": 1.0174021057470963,
+    "producer_headroom": 3.824478721590698,
     "sustainable": true,
-    "buffer_seconds": 60.160332397736504,
+    "buffer_seconds": 60.93952395987259,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 526532.9633623293,
-    "upload_bytes_per_second": 5298514.531116768,
+    "generation_bytes_per_second": 495736.6974019932,
+    "upload_bytes_per_second": 5307267.646444998,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 112.98171687748345,
+    "estimated_full_shards_per_hour": 106.37355510277602,
     "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": 1.1112560749988916,
-      "bytes": 2663601,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.182102421998934,
       "bytes": 5962190,
@@ -562,6 +556,12 @@
       "bytes": 6350565,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 0.9621634939976502,
+      "bytes": 2554887,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -100,3 +100,4 @@
 {"step": 290200, "loss": 0.5885871648788452, "lr": 0.0003, "grad_norm": 0.15245661774963712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42477362248, "tokens_per_second": 151496.21065290237, "data_wait_s": 1.3606000720756128e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 290300, "loss": 0.5866992063820362, "lr": 0.0003, "grad_norm": 0.19997807516574245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42491986897, "tokens_per_second": 150471.61368147776, "data_wait_s": 9.3019989435561e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 290400, "loss": 0.5906467884778976, "lr": 0.0003, "grad_norm": 0.15580875198805033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42506482545, "tokens_per_second": 148809.45715666597, "data_wait_s": 2.0689996745204553e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 290500, "loss": 0.5722408331930637, "lr": 0.0003, "grad_norm": 0.1425618393093551, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42520974684, "tokens_per_second": 149176.61279625515, "data_wait_s": 9.75399962044321e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4ca64e189892f61089a937b0d5d386f9eb377bc8bf68ae661b5d3c9fbf6b19db
-size 217767
+oid sha256:babef4cfd519f2d736c79089df6d84fb51598202e62520f7672b2ab20d45354e
+size 219885