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": 75223.16008289602,
-      "required_pages_per_second_upper_estimate": 0.005204112756313303
+      "required_token_positions_per_second": 74422.18355753737,
+      "required_pages_per_second_upper_estimate": 0.005148699341767399
     },
     {
       "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": 9402.895010362003,
-      "required_pages_per_second_upper_estimate": 0.003936795786062508
+      "required_token_positions_per_second": 9302.772944692171,
+      "required_pages_per_second_upper_estimate": 0.003894876767953032
     },
     {
       "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": 18805.790020724005,
-      "required_pages_per_second_upper_estimate": 0.003916750119283306
+      "required_token_positions_per_second": 18605.545889384342,
+      "required_pages_per_second_upper_estimate": 0.003875044547518116
     },
     {
       "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": 28208.68503108601,
-      "required_pages_per_second_upper_estimate": 0.0016910675954455233
+      "required_token_positions_per_second": 27908.318834076515,
+      "required_pages_per_second_upper_estimate": 0.0016730611005674311
     },
     {
       "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": 28208.68503108601,
-      "required_pages_per_second_upper_estimate": 0.0021931011842743356
+      "required_token_positions_per_second": 27908.318834076515,
+      "required_pages_per_second_upper_estimate": 0.0021697490336281223
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28208.68503108601,
-      "required_pages_per_second_upper_estimate": 0.36183072345257256
+      "required_token_positions_per_second": 27908.318834076515,
+      "required_pages_per_second_upper_estimate": 0.35797794838543007
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4050863042890569,
-    "producer_p95_seconds": 0.2823215030002757,
-    "gpu_update_seconds": 0.9801236735966907,
-    "producer_batches_per_second": 2.468609749112701,
-    "gpu_batches_per_second": 1.0202794065062937,
-    "producer_headroom": 2.4195428559769456,
+    "producer_batch_seconds": 0.07098568189846333,
+    "producer_p95_seconds": 0.2645800419995794,
+    "gpu_update_seconds": 0.9906723570264411,
+    "producer_batches_per_second": 14.087347944764163,
+    "gpu_batches_per_second": 1.0094154670889943,
+    "producer_headroom": 13.955946192691105,
     "sustainable": true,
-    "buffer_seconds": 60.767667762994826,
+    "buffer_seconds": 60.43101377861291,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 400229.5595867917,
-    "upload_bytes_per_second": 5231562.932004833,
+    "generation_bytes_per_second": 403331.9327030508,
+    "upload_bytes_per_second": 5315995.1859128475,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.87994662001432,
+    "estimated_full_shards_per_hour": 86.54564367121355,
     "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": 1.2962373409991415,
-      "bytes": 6282079,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.4015076579999004,
       "bytes": 7431338,
@@ -562,6 +556,12 @@
       "bytes": 989214,
       "pages": 29,
       "shards": 1
+    },
+    {
+      "seconds": 1.3733238169988908,
+      "bytes": 13560061,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -237,3 +237,4 @@
 {"step": 743700, "loss": 0.5052394866943359, "lr": 5e-05, "grad_norm": 0.2298700973910351, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108857464980, "tokens_per_second": 150381.66727331592, "data_wait_s": 1.0370997188147157e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 743800, "loss": 0.5453187227249146, "lr": 5e-05, "grad_norm": 0.22436211506852036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108871934168, "tokens_per_second": 147656.21708761793, "data_wait_s": 1.6111000149976462e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 743900, "loss": 0.5258931517601013, "lr": 5e-05, "grad_norm": 0.2305391180916694, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108886548112, "tokens_per_second": 150104.7790159702, "data_wait_s": 5.028300074627623e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 744000, "loss": 0.5613822340965271, "lr": 5e-05, "grad_norm": 0.23755027711396634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108901249779, "tokens_per_second": 151497.57460553377, "data_wait_s": 1.0911000572377816e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a440e11a10cd126f2192bbd030d0e2102df598d731573e289e5b9371c900b03b
-size 509499
+oid sha256:1e394d66996392a4416904725afa6151a64e214fa012b5d36f4ae4d3d68e7d45
+size 511617