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": 75450.07248263973,
-      "required_pages_per_second_upper_estimate": 0.006539821056334782
+      "required_token_positions_per_second": 74811.50791700376,
+      "required_pages_per_second_upper_estimate": 0.0064844718982125515
     },
     {
       "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": 9431.259060329967,
-      "required_pages_per_second_upper_estimate": 0.0037622702490545585
+      "required_token_positions_per_second": 9351.43848962547,
+      "required_pages_per_second_upper_estimate": 0.0037304286299766516
     },
     {
       "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": 18862.518120659934,
-      "required_pages_per_second_upper_estimate": 0.0043635115982339
+      "required_token_positions_per_second": 18702.87697925094,
+      "required_pages_per_second_upper_estimate": 0.004326581429757059
     },
     {
       "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": 28293.777180989902,
-      "required_pages_per_second_upper_estimate": 0.001722942584345926
+      "required_token_positions_per_second": 28054.31546887641,
+      "required_pages_per_second_upper_estimate": 0.0017083606224367212
     },
     {
       "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": 28293.777180989902,
-      "required_pages_per_second_upper_estimate": 0.002178133289580683
+      "required_token_positions_per_second": 28054.31546887641,
+      "required_pages_per_second_upper_estimate": 0.0021596988641097415
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28293.777180989902,
-      "required_pages_per_second_upper_estimate": 0.3513140194071036
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28054.31546887641,
+      "required_pages_per_second_upper_estimate": 0.3483407063694502
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1407589353672023,
-    "producer_p95_seconds": 0.2633858820008754,
-    "gpu_update_seconds": 0.9771759996249709,
-    "producer_batches_per_second": 7.104344725194662,
-    "gpu_batches_per_second": 1.0233571029003872,
-    "producer_headroom": 6.942195158522483,
+    "producer_batch_seconds": 0.43249772474213444,
+    "producer_p95_seconds": 0.2588358339999104,
+    "gpu_update_seconds": 0.9855168282638306,
+    "producer_batches_per_second": 2.312150891883244,
+    "gpu_batches_per_second": 1.0146960166694303,
+    "producer_headroom": 2.2786636134361618,
     "sustainable": true,
-    "buffer_seconds": 60.584911976748195,
+    "buffer_seconds": 60.116526524093665,
     "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": 221782.54867383945,
-    "upload_bytes_per_second": 6554691.028003118,
+    "generation_bytes_per_second": 210233.97680507918,
+    "upload_bytes_per_second": 6532109.509576638,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 47.58937211190594,
+    "estimated_full_shards_per_hour": 45.111317425864044,
     "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": 1.260473302999344,
-      "bytes": 10994776,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 0.9115331740003967,
       "bytes": 8005286,
@@ -562,6 +556,12 @@
       "bytes": 5888539,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.030597670998759,
+      "bytes": 7980071,
+      "pages": 26,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -118,3 +118,4 @@
 {"step": 11800, "loss": 0.7083174623548985, "lr": 0.0003, "grad_norm": 0.15130096636100004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1725960527, "tokens_per_second": 148386.64885272944, "data_wait_s": 1.5227000403683633e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 11900, "loss": 0.6477013938128948, "lr": 0.0003, "grad_norm": 0.1480410445894246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1740609258, "tokens_per_second": 110180.41730813945, "data_wait_s": 1.3979000868857838e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 12000, "loss": 0.7293434552848339, "lr": 0.0003, "grad_norm": 0.15609938118390926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1755365069, "tokens_per_second": 150381.96532769338, "data_wait_s": 1.2484000762924552e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 12100, "loss": 0.7310982514172792, "lr": 0.0003, "grad_norm": 0.15563355228699358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1770075350, "tokens_per_second": 149302.15500441284, "data_wait_s": 8.760000127949752e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c1a1795670c711b14656e68f68756ac671b7f5397120560342250c1237ce1bc2
-size 251632
+oid sha256:f65590a3a4546e7ed6cf136c7cbce16584b7d8c3d163c41a4ce7bc7eaf478c88
+size 253715