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": 75241.33181577931,
-      "required_pages_per_second_upper_estimate": 0.006521727944379704
+      "required_token_positions_per_second": 75452.27296919293,
+      "required_pages_per_second_upper_estimate": 0.006540011788932131
     },
     {
       "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": 9405.166476972414,
-      "required_pages_per_second_upper_estimate": 0.003751861527434344
+      "required_token_positions_per_second": 9431.534121149116,
+      "required_pages_per_second_upper_estimate": 0.0037623799749278426
     },
     {
       "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": 18810.332953944828,
-      "required_pages_per_second_upper_estimate": 0.004351439478341927
+      "required_token_positions_per_second": 18863.06824229823,
+      "required_pages_per_second_upper_estimate": 0.00436363885919311
     },
     {
       "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": 28215.49943091724,
-      "required_pages_per_second_upper_estimate": 0.0017181758800581155
+      "required_token_positions_per_second": 28294.60236344735,
+      "required_pages_per_second_upper_estimate": 0.0017229928336281867
     },
     {
       "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": 28215.49943091724,
-      "required_pages_per_second_upper_estimate": 0.0021721072516934082
+      "required_token_positions_per_second": 28294.60236344735,
+      "required_pages_per_second_upper_estimate": 0.002178196814410508
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28215.49943091724,
-      "required_pages_per_second_upper_estimate": 0.350342071729978
+      "required_token_positions_per_second": 28294.60236344735,
+      "required_pages_per_second_upper_estimate": 0.35132426541151707
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3953732031797301,
-    "producer_p95_seconds": 0.3005809370006318,
-    "gpu_update_seconds": 0.9798869613381572,
-    "producer_batches_per_second": 2.529255882689188,
-    "gpu_batches_per_second": 1.020525876407597,
-    "producer_headroom": 2.478384861334967,
+    "producer_batch_seconds": 0.12159875697673783,
+    "producer_p95_seconds": 0.31615101199713536,
+    "gpu_update_seconds": 0.9771475013099082,
+    "producer_batches_per_second": 8.223768275783467,
+    "gpu_batches_per_second": 1.0233869489094092,
+    "producer_headroom": 8.035834622033507,
     "sustainable": true,
-    "buffer_seconds": 60.75299160296575,
+    "buffer_seconds": 60.58314508121431,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 262957.2553526859,
-    "upload_bytes_per_second": 5577137.306538899,
+    "generation_bytes_per_second": 274623.518683876,
+    "upload_bytes_per_second": 5538647.937039794,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.424505667070704,
+    "estimated_full_shards_per_hour": 58.92781420123301,
     "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": 1.255419142999017,
-      "bytes": 11119971,
-      "pages": 21,
-      "shards": 3
-    },
     {
       "seconds": 1.229843424000137,
       "bytes": 7832261,
@@ -562,6 +556,12 @@
       "bytes": 7833557,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.04416848999972,
+      "bytes": 7007035,
+      "pages": 42,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -201,3 +201,4 @@
 {"step": 20100, "loss": 0.6706032436341047, "lr": 0.0003, "grad_norm": 0.13800229756680835, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2941240165, "tokens_per_second": 134596.65403390976, "data_wait_s": 1.006900129141286e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 20200, "loss": 0.6517019756138325, "lr": 0.0003, "grad_norm": 0.14619521083157333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2955752885, "tokens_per_second": 124759.274725961, "data_wait_s": 1.484200038248673e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 20300, "loss": 0.6576510518789291, "lr": 0.0003, "grad_norm": 0.13662180698471504, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2970333450, "tokens_per_second": 147303.99292384667, "data_wait_s": 1.3060998753644526e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 20400, "loss": 0.6723321489989758, "lr": 0.0003, "grad_norm": 0.13572539836841088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2984949151, "tokens_per_second": 148670.7590633348, "data_wait_s": 1.4277000445872545e-05, "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 sha256:79b90c02172ac5c65ded2476b316cd45176ad8d2353d3946f9e8fee1434e86b8
-size 427477
+oid sha256:d0b949665a3579d1d9c4f097f91ebe72b852fe38147e5d22860c538e52b8ed96
+size 429595