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": 74476.83668285204,
-      "required_pages_per_second_upper_estimate": 0.006455463443852768
+      "required_token_positions_per_second": 74347.6496256331,
+      "required_pages_per_second_upper_estimate": 0.006444265837154634
     },
     {
       "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": 9309.604585356505,
-      "required_pages_per_second_upper_estimate": 0.003713740460091154
+      "required_token_positions_per_second": 9293.456203204138,
+      "required_pages_per_second_upper_estimate": 0.0037072986290107458
     },
     {
       "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": 18619.20917071301,
-      "required_pages_per_second_upper_estimate": 0.004307226354754947
+      "required_token_positions_per_second": 18586.912406408275,
+      "required_pages_per_second_upper_estimate": 0.004299755066736685
     },
     {
       "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": 27928.813756069518,
-      "required_pages_per_second_upper_estimate": 0.0017007182265833805
+      "required_token_positions_per_second": 27880.368609612415,
+      "required_pages_per_second_upper_estimate": 0.0016977681713361828
     },
     {
       "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": 27928.813756069518,
-      "required_pages_per_second_upper_estimate": 0.0021500373948468905
+      "required_token_positions_per_second": 27880.368609612415,
+      "required_pages_per_second_upper_estimate": 0.002146307953367874
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27928.813756069518,
-      "required_pages_per_second_upper_estimate": 0.3467823951236018
+      "required_token_positions_per_second": 27880.368609612415,
+      "required_pages_per_second_upper_estimate": 0.3461808685400799
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10318995754686,
-    "producer_p95_seconds": 0.22641479300000356,
-    "gpu_update_seconds": 0.9899453747473077,
-    "producer_batches_per_second": 9.690865504483671,
-    "gpu_batches_per_second": 1.0101567475430235,
-    "producer_headroom": 9.593427483461845,
+    "producer_batch_seconds": 0.08201863790631592,
+    "producer_p95_seconds": 0.20107217100121488,
+    "gpu_update_seconds": 0.9916655115695888,
+    "producer_batches_per_second": 12.19235073303984,
+    "gpu_batches_per_second": 1.008404535937949,
+    "producer_headroom": 12.090733726915804,
     "sustainable": true,
-    "buffer_seconds": 60.38666785958577,
+    "buffer_seconds": 60.49159620574492,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 211481.5878781797,
+    "generation_bytes_per_second": 203079.3717311272,
     "upload_bytes_per_second": 6572480.029369081,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 45.37902571925205,
+    "estimated_full_shards_per_hour": 43.57610572767603,
     "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": [

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

@@ -92,3 +92,4 @@
 {"step": 9200, "loss": 0.8535147793591022, "lr": 0.0003, "grad_norm": 0.33993378266042257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1345371939, "tokens_per_second": 150043.87223557767, "data_wait_s": 1.4172999726724811e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 9300, "loss": 0.7113298326730728, "lr": 0.0003, "grad_norm": 0.1712786023034142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1360127552, "tokens_per_second": 150064.52455482446, "data_wait_s": 1.2900998626719229e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 9400, "loss": 0.7275722995400429, "lr": 0.0003, "grad_norm": 0.15210553956700304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1374753001, "tokens_per_second": 140530.373021818, "data_wait_s": 1.2996000805287622e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 9500, "loss": 0.7265461198985577, "lr": 0.0003, "grad_norm": 0.17726367278197366, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1389271585, "tokens_per_second": 147123.0570736442, "data_wait_s": 9.096000212593935e-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 sha256:5969f5515294cd8f73fc58f08a520959b3bd06a45c7c2bf07a42d165305f5d22
-size 196701
+oid sha256:85c1b9cff0910b0202ef28bb00739ca612d3efd585e8aff019271971891dfe4a
+size 198784