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": 76988.93247809267,
-      "required_pages_per_second_upper_estimate": 0.005326272987769443
+      "required_token_positions_per_second": 77068.33740044276,
+      "required_pages_per_second_upper_estimate": 0.0053317664045424255
     },
     {
       "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": 9623.616559761584,
-      "required_pages_per_second_upper_estimate": 0.004029207289606033
+      "required_token_positions_per_second": 9633.542175055345,
+      "required_pages_per_second_upper_estimate": 0.004033362937459114
     },
     {
       "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": 19247.233119523167,
-      "required_pages_per_second_upper_estimate": 0.00400869107512579
+      "required_token_positions_per_second": 19267.08435011069,
+      "required_pages_per_second_upper_estimate": 0.004012825562945016
     },
     {
       "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": 28870.849679284747,
-      "required_pages_per_second_upper_estimate": 0.0017307633550381625
+      "required_token_positions_per_second": 28900.626525166033,
+      "required_pages_per_second_upper_estimate": 0.0017325484314821345
     },
     {
       "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": 28870.849679284747,
-      "required_pages_per_second_upper_estimate": 0.0022445815731173082
+      "required_token_positions_per_second": 28900.626525166033,
+      "required_pages_per_second_upper_estimate": 0.002246896591910075
     },
     {
       "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": 28870.849679284747,
-      "required_pages_per_second_upper_estimate": 0.3703242605826599
+      "required_token_positions_per_second": 28900.626525166033,
+      "required_pages_per_second_upper_estimate": 0.3707062059897389
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07183294161717768,
-    "producer_p95_seconds": 0.14796916099976443,
-    "gpu_update_seconds": 0.9576441395778469,
-    "producer_batches_per_second": 13.921189603084084,
-    "gpu_batches_per_second": 1.0442292274046856,
-    "producer_headroom": 13.331545639345526,
+    "producer_batch_seconds": 0.06276294911717173,
+    "producer_p95_seconds": 0.15694820800035814,
+    "gpu_update_seconds": 0.9566574612465486,
+    "producer_batches_per_second": 15.932967046100824,
+    "gpu_batches_per_second": 1.0453062255919428,
+    "producer_headroom": 15.242391804447735,
     "sustainable": true,
-    "buffer_seconds": 60.331580793404356,
+    "buffer_seconds": 60.26942005853256,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 271572.56908326835,
-    "upload_bytes_per_second": 5306715.561505132,
+    "generation_bytes_per_second": 296501.8517850054,
+    "upload_bytes_per_second": 5299760.673765951,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.273151439414384,
+    "estimated_full_shards_per_hour": 63.62239518320677,
     "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": [
@@ -358,6 +358,12 @@
       "bytes": 2658357,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1752439570000206,
+      "bytes": 5996670,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -33,3 +33,4 @@
 {"step": 400400, "loss": 0.6056730151176453, "lr": 0.0003, "grad_norm": 0.16038392755760572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58607306918, "tokens_per_second": 151749.90916819338, "data_wait_s": 6.699000095977681e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 400500, "loss": 0.5657328367233276, "lr": 0.0003, "grad_norm": 0.1569044007497574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58621964614, "tokens_per_second": 152573.03605078027, "data_wait_s": 9.83200015980401e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 400600, "loss": 0.5313172936439514, "lr": 0.0003, "grad_norm": 0.14731284188264282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58636565138, "tokens_per_second": 151862.11140838277, "data_wait_s": 7.24300025467528e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 400700, "loss": 0.6032090187072754, "lr": 0.0003, "grad_norm": 0.1455071352752887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58651151433, "tokens_per_second": 151886.13541794656, "data_wait_s": 1.004999967335607e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f12060f63799f6f5e3a16f77bdb5f3f9e705712daa56d9311193463d13b8c280
-size 75078
+oid sha256:007be3c05973f2837602b07c2ac9c99b06a78b8817f23b0ea6510248f1be59bd
+size 77196