Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77067.21685127672,
-      "required_pages_per_second_upper_estimate": 0.0053316888823510865
+      "required_token_positions_per_second": 77367.02049284047,
+      "required_pages_per_second_upper_estimate": 0.005352429993914756
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9633.40210640959,
-      "required_pages_per_second_upper_estimate": 0.0040333042936421025
+      "required_token_positions_per_second": 9670.877561605059,
+      "required_pages_per_second_upper_estimate": 0.004048994484155951
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19266.80421281918,
-      "required_pages_per_second_upper_estimate": 0.004012767217734912
+      "required_token_positions_per_second": 19341.755123210118,
+      "required_pages_per_second_upper_estimate": 0.004028377515780918
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28900.20631922877,
-      "required_pages_per_second_upper_estimate": 0.001732523240777813
+      "required_token_positions_per_second": 29012.632684815173,
+      "required_pages_per_second_upper_estimate": 0.0017392630297295973
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28900.20631922877,
-      "required_pages_per_second_upper_estimate": 0.0022468639227467436
+      "required_token_positions_per_second": 29012.632684815173,
+      "required_pages_per_second_upper_estimate": 0.0022556045781597658
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28900.20631922877,
-      "required_pages_per_second_upper_estimate": 0.37070081603915767
+      "required_token_positions_per_second": 29012.632684815173,
+      "required_pages_per_second_upper_estimate": 0.37214290074287365
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06533094089826363,
-    "producer_p95_seconds": 0.22478549700099393,
-    "gpu_update_seconds": 0.956671370944656,
-    "producer_batches_per_second": 15.30668296293553,
-    "gpu_batches_per_second": 1.0452910271711795,
-    "producer_headroom": 14.643465374766743,
+    "producer_batch_seconds": 0.29298818198427057,
+    "producer_p95_seconds": 0.2577388919999066,
+    "gpu_update_seconds": 0.9529641897844933,
+    "producer_batches_per_second": 3.4131069493229123,
+    "gpu_batches_per_second": 1.0493573743061044,
+    "producer_headroom": 3.2525686986093327,
     "sustainable": true,
-    "buffer_seconds": 60.27029636951333,
+    "buffer_seconds": 60.036743956423074,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 406124.8569051563,
-    "upload_bytes_per_second": 5796207.590347337,
+    "generation_bytes_per_second": 393600.9617299802,
+    "upload_bytes_per_second": 5733009.482123485,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.14494018903748,
+    "estimated_full_shards_per_hour": 84.45760382580332,
     "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": 1.1898289029995794,
-      "bytes": 7719194,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.9053547809999145,
       "bytes": 813842,
@@ -562,6 +556,12 @@
       "bytes": 6175250,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.8423896300009801,
+      "bytes": 1034182,
+      "pages": 31,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -204,3 +204,4 @@
 {"step": 705400, "loss": 0.617925226688385, "lr": 8e-05, "grad_norm": 0.23067745386033775, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103250575496, "tokens_per_second": 151865.75045943816, "data_wait_s": 1.5477999113500118e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 705500, "loss": 0.5783052444458008, "lr": 8e-05, "grad_norm": 0.21751230000714367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103265291575, "tokens_per_second": 153680.89974700272, "data_wait_s": 7.798000297043473e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 705600, "loss": 0.5398990511894226, "lr": 8e-05, "grad_norm": 0.2059432494326504, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103279953452, "tokens_per_second": 152517.2289203496, "data_wait_s": 9.87699968391098e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 705700, "loss": 0.5683505535125732, "lr": 8e-05, "grad_norm": 0.2149800912421205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103294723259, "tokens_per_second": 153518.86430251857, "data_wait_s": 1.621000046725385e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fd94ea36b676c22bc475749b591098e356b2eb4fe1d5f0b3347d64a375014787
-size 439605
+oid sha256:0486da36c03c4458bb6ce2d6356fdac5cca7d8a5ab232bb4bc52f36f8f0d00ca
+size 441723