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": 75363.06027600405,
-      "required_pages_per_second_upper_estimate": 0.006532279058794023
+      "required_token_positions_per_second": 75234.4902396709,
+      "required_pages_per_second_upper_estimate": 0.006521134933902443
     },
     {
       "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": 9420.382534500506,
-      "required_pages_per_second_upper_estimate": 0.0037579314402826336
+      "required_token_positions_per_second": 9404.311279958862,
+      "required_pages_per_second_upper_estimate": 0.003751520376559303
     },
     {
       "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": 18840.765069001012,
-      "required_pages_per_second_upper_estimate": 0.004358479412573243
+      "required_token_positions_per_second": 18808.622559917723,
+      "required_pages_per_second_upper_estimate": 0.004351043808785663
     },
     {
       "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": 28261.147603501522,
-      "required_pages_per_second_upper_estimate": 0.001720955614271046
+      "required_token_positions_per_second": 28212.933839876587,
+      "required_pages_per_second_upper_estimate": 0.0017180196490243556
     },
     {
       "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": 28261.147603501522,
-      "required_pages_per_second_upper_estimate": 0.0021756213743811758
+      "required_token_positions_per_second": 28212.933839876587,
+      "required_pages_per_second_upper_estimate": 0.0021719097453930885
     },
     {
       "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": 28261.147603501522,
-      "required_pages_per_second_upper_estimate": 0.3509088692588689
+      "required_token_positions_per_second": 28212.933839876587,
+      "required_pages_per_second_upper_estimate": 0.35031021567573395
     }
   ],
   "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": 0.2908222494845063,
-    "producer_p95_seconds": 0.3258510430023307,
-    "gpu_update_seconds": 0.9783042213251966,
-    "producer_batches_per_second": 3.4385264599683785,
-    "gpu_batches_per_second": 1.0221769243164613,
-    "producer_headroom": 3.363924950925449,
+    "producer_batch_seconds": 0.10071946053895431,
+    "producer_p95_seconds": 0.352619577999576,
+    "gpu_update_seconds": 0.9799760690227083,
+    "producer_batches_per_second": 9.92856787207711,
+    "gpu_batches_per_second": 1.0204330815927585,
+    "producer_headroom": 9.729758914303282,
     "sustainable": true,
-    "buffer_seconds": 60.65486172216219,
+    "buffer_seconds": 60.758516279407914,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 433485.73352486925,
-    "upload_bytes_per_second": 5768313.589890935,
+    "generation_bytes_per_second": 390367.3418064342,
+    "upload_bytes_per_second": 5904023.614435943,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.0159473830181,
+    "estimated_full_shards_per_hour": 83.76374426502962,
     "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": 1.0657177410030272,
-      "bytes": 1129396,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.0460155590008071,
       "bytes": 6970566,
@@ -562,6 +556,12 @@
       "bytes": 7349596,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.122332668001036,
+      "bytes": 11201873,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -268,3 +268,4 @@
 {"step": 26800, "loss": 0.6303611919283867, "lr": 0.0003, "grad_norm": 0.13173461040175014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3921706348, "tokens_per_second": 148180.66551602937, "data_wait_s": 1.195900040329434e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 26900, "loss": 0.7829932682216167, "lr": 0.0003, "grad_norm": 0.15855550865740187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3936429389, "tokens_per_second": 148950.4094190557, "data_wait_s": 1.2045002222293988e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 27000, "loss": 0.6335824616253376, "lr": 0.0003, "grad_norm": 0.1356002792379001, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3951104300, "tokens_per_second": 149903.06616645158, "data_wait_s": 1.081600203178823e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 27100, "loss": 0.6737250126898289, "lr": 0.0003, "grad_norm": 0.13846469064501848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3965767704, "tokens_per_second": 136907.83867136188, "data_wait_s": 1.0435000149300322e-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:ee7a275663d2644d2cad2351893b3b919e5af98d1ebcdcbdb4522700d5b160ba
-size 570166
+oid sha256:42d27c4d8ce8408af9f0bd6eb6e920218b9774873360dbd54a1d234f0e5cb354
+size 572284