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": 75954.62140342173,
-      "required_pages_per_second_upper_estimate": 0.006583554078020617
+      "required_token_positions_per_second": 73888.83444313298,
+      "required_pages_per_second_upper_estimate": 0.0064044968999919715
     },
     {
       "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": 9494.327675427716,
-      "required_pages_per_second_upper_estimate": 0.003787429262576877
+      "required_token_positions_per_second": 9236.104305391622,
+      "required_pages_per_second_upper_estimate": 0.003684420099486047
     },
     {
       "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": 18988.65535085543,
-      "required_pages_per_second_upper_estimate": 0.004392691226500198
+      "required_token_positions_per_second": 18472.208610783244,
+      "required_pages_per_second_upper_estimate": 0.00427322036233669
     },
     {
       "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": 28482.983026283146,
-      "required_pages_per_second_upper_estimate": 0.001734464227637933
+      "required_token_positions_per_second": 27708.312916174866,
+      "required_pages_per_second_upper_estimate": 0.0016872908823122948
     },
     {
       "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": 28482.983026283146,
-      "required_pages_per_second_upper_estimate": 0.002192698879306658
+      "required_token_positions_per_second": 27708.312916174866,
+      "required_pages_per_second_upper_estimate": 0.0021330626298064087
     },
     {
       "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": 28482.983026283146,
-      "required_pages_per_second_upper_estimate": 0.35366332277441603
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27708.312916174866,
+      "required_pages_per_second_upper_estimate": 0.3440445126609492
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.10889414641397366,
-    "producer_p95_seconds": 0.2926854949982953,
-    "gpu_update_seconds": 0.970684846263727,
-    "producer_batches_per_second": 9.183230071875348,
-    "gpu_batches_per_second": 1.0302004856149864,
-    "producer_headroom": 8.914022270522757,
+    "producer_batch_seconds": 0.08355192261697653,
+    "producer_p95_seconds": 0.3432925699999032,
+    "gpu_update_seconds": 0.9978232916469029,
+    "producer_batches_per_second": 11.96860549318843,
+    "gpu_batches_per_second": 1.002181456748223,
+    "producer_headroom": 11.942553329636484,
     "sustainable": true,
-    "buffer_seconds": 60.182460468351074,
+    "buffer_seconds": 60.867220790461076,
     "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": 361262.32990183355,
-    "upload_bytes_per_second": 4081624.038984109,
+    "generation_bytes_per_second": 315760.0117863258,
+    "upload_bytes_per_second": 4364849.523467122,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.51848624030356,
+    "estimated_full_shards_per_hour": 67.75474801246958,
     "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": 8.626221913000336,
-      "bytes": 11949436,
-      "pages": 25,
-      "shards": 4
-    },
     {
       "seconds": 13.160608485999546,
       "bytes": 6360120,
@@ -562,6 +556,12 @@
       "bytes": 2567688,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.52958795199811,
+      "bytes": 10493875,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -240,3 +240,4 @@
 {"step": 304200, "loss": 0.5576318129897118, "lr": 0.0003, "grad_norm": 0.1558965124334837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44526920027, "tokens_per_second": 148738.75476426695, "data_wait_s": 1.0335999832022935e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 304300, "loss": 0.58489940315485, "lr": 0.0003, "grad_norm": 0.19894666184799137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44541510185, "tokens_per_second": 149116.8349708186, "data_wait_s": 9.79400283540599e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 304400, "loss": 0.5945089161396027, "lr": 0.0003, "grad_norm": 0.14332697843830636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44556189442, "tokens_per_second": 141899.99781833962, "data_wait_s": 1.3651999324792996e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 304500, "loss": 0.6310973726212978, "lr": 0.0003, "grad_norm": 0.16308966050276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44570812241, "tokens_per_second": 149366.05639460578, "data_wait_s": 1.0029001714428887e-05, "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 sha256:1f8292803b0563760327abf86995801b1570eeecb55ac7155b610b3cddca461e
-size 515853
+oid sha256:be01466bd0ba21f0e596c584e6de6432a75e4b73482cc0da5ad27495189bbdf1
+size 517971