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": 77399.81225310065,
-      "required_pages_per_second_upper_estimate": 0.005354698603976921
+      "required_token_positions_per_second": 77222.9557560021,
+      "required_pages_per_second_upper_estimate": 0.005342463261144034
     },
     {
       "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": 9674.976531637582,
-      "required_pages_per_second_upper_estimate": 0.00405071063731234
+      "required_token_positions_per_second": 9652.869469500263,
+      "required_pages_per_second_upper_estimate": 0.004041454872043398
     },
     {
       "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": 19349.953063275163,
-      "required_pages_per_second_upper_estimate": 0.004030084930502248
+      "required_token_positions_per_second": 19305.738939000526,
+      "required_pages_per_second_upper_estimate": 0.004020876294420709
     },
     {
       "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": 29024.929594912745,
-      "required_pages_per_second_upper_estimate": 0.001740000210713652
+      "required_token_positions_per_second": 28958.60840850079,
+      "required_pages_per_second_upper_estimate": 0.0017360243568548435
     },
     {
       "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": 29024.929594912745,
-      "required_pages_per_second_upper_estimate": 0.002256560608831461
+      "required_token_positions_per_second": 28958.60840850079,
+      "required_pages_per_second_upper_estimate": 0.0022514044283040006
     },
     {
       "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": 29024.929594912745,
-      "required_pages_per_second_upper_estimate": 0.37230063230221194
+      "required_token_positions_per_second": 28958.60840850079,
+      "required_pages_per_second_upper_estimate": 0.37144993533306125
     }
   ],
   "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": 0.08073031621879068,
-    "producer_p95_seconds": 0.2216414070007886,
-    "gpu_update_seconds": 0.9525604501326996,
-    "producer_batches_per_second": 12.386920389235902,
-    "gpu_batches_per_second": 1.049802141019703,
-    "producer_headroom": 11.799290461728464,
+    "producer_batch_seconds": 0.2416896958671373,
+    "producer_p95_seconds": 0.22406062499976542,
+    "gpu_update_seconds": 0.954742010043685,
+    "producer_batches_per_second": 4.137536755185973,
+    "gpu_batches_per_second": 1.047403371256539,
+    "producer_headroom": 3.9502801582758824,
     "sustainable": true,
-    "buffer_seconds": 60.011308358360075,
+    "buffer_seconds": 60.14874663275215,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 412752.5019289825,
-    "upload_bytes_per_second": 5604774.308749773,
+    "generation_bytes_per_second": 402821.20529372257,
+    "upload_bytes_per_second": 5622343.015514161,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.56707852747064,
+    "estimated_full_shards_per_hour": 86.43605345829732,
     "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": 1.1581361090002247,
-      "bytes": 10067921,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.7976953110000977,
       "bytes": 1131427,
@@ -562,6 +556,12 @@
       "bytes": 11950097,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.2208401020016026,
+      "bytes": 11777262,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -103,3 +103,4 @@
 {"step": 567400, "loss": 0.5477515459060669, "lr": 0.0003, "grad_norm": 0.16141831762377298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83050252354, "tokens_per_second": 153355.19655005092, "data_wait_s": 1.1143998563056812e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 567500, "loss": 0.565815806388855, "lr": 0.0003, "grad_norm": 0.1765819460917496, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83064988208, "tokens_per_second": 153172.1090037539, "data_wait_s": 1.0559000656940043e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 567600, "loss": 0.6129605174064636, "lr": 0.0003, "grad_norm": 0.15824549312024988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83079676448, "tokens_per_second": 153325.14696044597, "data_wait_s": 1.771099960024003e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 567700, "loss": 0.5564883351325989, "lr": 0.0003, "grad_norm": 0.1546044285232266, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83094420347, "tokens_per_second": 153465.20481614032, "data_wait_s": 8.302999049192294e-06, "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 sha256:4c6f4fbb0b92e0da3958e15460876e5fd8f15489d1c2c6f716f0c317f289ab64
-size 224121
+oid sha256:8f65d673b0fa174e34fa199c37ed7e90600a08fc57ccab760b089874aa2a9537
+size 226239