Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77154.22677465378,
-      "required_pages_per_second_upper_estimate": 0.005337708430741149
+      "required_token_positions_per_second": 76755.01463057223,
+      "required_pages_per_second_upper_estimate": 0.005310090008313794
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9644.278346831723,
-      "required_pages_per_second_upper_estimate": 0.004037857948385123
+      "required_token_positions_per_second": 9594.376828821529,
+      "required_pages_per_second_upper_estimate": 0.004016965224856447
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19288.556693663446,
-      "required_pages_per_second_upper_estimate": 0.004017297685843276
+      "required_token_positions_per_second": 19188.753657643058,
+      "required_pages_per_second_upper_estimate": 0.003996511345423282
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28932.83504049517,
-      "required_pages_per_second_upper_estimate": 0.0017344792827965608
+      "required_token_positions_per_second": 28783.130486464586,
+      "required_pages_per_second_upper_estimate": 0.001725504723368046
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28932.83504049517,
-      "required_pages_per_second_upper_estimate": 0.0022494006623066287
+      "required_token_positions_per_second": 28783.130486464586,
+      "required_pages_per_second_upper_estimate": 0.0022377617917114937
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28932.83504049517,
-      "required_pages_per_second_upper_estimate": 0.37111934224157167
+      "required_token_positions_per_second": 28783.130486464586,
+      "required_pages_per_second_upper_estimate": 0.369199092962694
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08967404627338738,
-    "producer_p95_seconds": 0.2518267619998369,
-    "gpu_update_seconds": 0.955592494178435,
-    "producer_batches_per_second": 11.151498583562528,
-    "gpu_batches_per_second": 1.0464711747864284,
-    "producer_headroom": 10.656288345293802,
+    "producer_batch_seconds": 0.0734531932658058,
+    "producer_p95_seconds": 0.26069440400169697,
+    "gpu_update_seconds": 0.9605626466864545,
+    "producer_batches_per_second": 13.614112001656485,
+    "gpu_batches_per_second": 1.0410565135440026,
+    "producer_headroom": 13.077207456596978,
     "sustainable": true,
-    "buffer_seconds": 60.202327133241404,
+    "buffer_seconds": 60.515446741246635,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344485.0152584313,
-    "upload_bytes_per_second": 5471685.619773158,
+    "generation_bytes_per_second": 343782.8517269979,
+    "upload_bytes_per_second": 5358563.892854183,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.91846507372574,
+    "estimated_full_shards_per_hour": 73.76779712541058,
     "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": 0.8145781550010724,
-      "bytes": 930729,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.3511633730013273,
       "bytes": 6255780,
@@ -562,6 +556,12 @@
       "bytes": 7147043,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 3.4036351569993712,
+      "bytes": 6159372,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -242,3 +242,4 @@
 {"step": 581300, "loss": 0.5507333278656006, "lr": 0.0003, "grad_norm": 0.16153292466350272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85085361082, "tokens_per_second": 152954.14175127208, "data_wait_s": 1.0825999197550118e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 581400, "loss": 0.5653789043426514, "lr": 0.0003, "grad_norm": 0.1666213727706197, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85100087281, "tokens_per_second": 153145.8002052712, "data_wait_s": 1.8106999050360173e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 581500, "loss": 0.5811357498168945, "lr": 0.0003, "grad_norm": 0.14319670271985702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85114845007, "tokens_per_second": 153675.46980002642, "data_wait_s": 9.942999895429239e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 581600, "loss": 0.6176038384437561, "lr": 0.0003, "grad_norm": 0.16526034513122184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85129605826, "tokens_per_second": 153454.87485712368, "data_wait_s": 1.1123996955575421e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d88295938428008753ab4bd02914aa742a56c1e8f7937910932d72cbf2b94c51
-size 520872
+oid sha256:cbc33baa20c4ee78b84941cb59a014cf334fd10f48e77326c3654045b8ac25ff
+size 522990