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": 76182.42988643878,
-      "required_pages_per_second_upper_estimate": 0.005270477267134993
+      "required_token_positions_per_second": 76088.07182957386,
+      "required_pages_per_second_upper_estimate": 0.005263949357820222
     },
     {
       "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": 9522.803735804848,
-      "required_pages_per_second_upper_estimate": 0.003986999065426503
+      "required_token_positions_per_second": 9511.008978696733,
+      "required_pages_per_second_upper_estimate": 0.003982060846927872
     },
     {
       "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": 19045.607471609695,
-      "required_pages_per_second_upper_estimate": 0.003966697769891317
+      "required_token_positions_per_second": 19022.017957393466,
+      "required_pages_per_second_upper_estimate": 0.003961784696177401
     },
     {
       "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": 28568.41120741454,
-      "required_pages_per_second_upper_estimate": 0.0017126326304463513
+      "required_token_positions_per_second": 28533.0269360902,
+      "required_pages_per_second_upper_estimate": 0.0017105113974091109
     },
     {
       "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": 28568.41120741454,
-      "required_pages_per_second_upper_estimate": 0.002221068312215649
+      "required_token_positions_per_second": 28533.0269360902,
+      "required_pages_per_second_upper_estimate": 0.0022183173407591423
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28568.41120741454,
-      "required_pages_per_second_upper_estimate": 0.3664449045986396
+      "required_token_positions_per_second": 28533.0269360902,
+      "required_pages_per_second_upper_estimate": 0.3659910331587614
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.17731562778149623,
-    "producer_p95_seconds": 0.26181521299804444,
-    "gpu_update_seconds": 0.9677822052919883,
-    "producer_batches_per_second": 5.63966082691982,
-    "gpu_batches_per_second": 1.0332903359163246,
-    "producer_headroom": 5.4579633921753015,
+    "producer_batch_seconds": 0.08705522939055754,
+    "producer_p95_seconds": 0.25625118700190797,
+    "gpu_update_seconds": 0.9689823677637662,
+    "producer_batches_per_second": 11.486960714487132,
+    "gpu_batches_per_second": 1.0320105228620586,
+    "producer_headroom": 11.130662391533104,
     "sustainable": true,
-    "buffer_seconds": 60.002496728103274,
+    "buffer_seconds": 60.07690680135351,
     "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": 352146.538909456,
-    "upload_bytes_per_second": 3782999.478219358,
+    "generation_bytes_per_second": 366813.70793951483,
+    "upload_bytes_per_second": 3806482.001671132,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.56244969809303,
+    "estimated_full_shards_per_hour": 78.7096827377232,
     "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": 1.317758003999188,
-      "bytes": 10010526,
-      "pages": 18,
-      "shards": 3
-    },
     {
       "seconds": 1.4487697540007503,
       "bytes": 2537038,
@@ -562,6 +556,12 @@
       "bytes": 4607080,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.4511087869977928,
+      "bytes": 13183448,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -229,3 +229,4 @@
 {"step": 667900, "loss": 0.6122384071350098, "lr": 0.0001, "grad_norm": 0.19803908538331397, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97763835607, "tokens_per_second": 131008.13372610754, "data_wait_s": 1.2407999747665599e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 668000, "loss": 0.48959314823150635, "lr": 0.0001, "grad_norm": 0.18453661882029934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97778281404, "tokens_per_second": 150372.9252406844, "data_wait_s": 6.645997927989811e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 668100, "loss": 0.5336928963661194, "lr": 0.0001, "grad_norm": 0.19112198194547503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97792781550, "tokens_per_second": 151095.12986784498, "data_wait_s": 1.194099968415685e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 668200, "loss": 0.5407228469848633, "lr": 0.0001, "grad_norm": 0.19003478777965516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97807303839, "tokens_per_second": 150074.80174617, "data_wait_s": 7.735998224234208e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:74c78bf4ccf1049229e12943b1430f6c3f7be8d16692695da105ca5a25447c30
-size 492555
+oid sha256:17b175cbc0ac80d270551c98df01629bb3cc9ccce18db6152c4d51eeaf05de5f
+size 494673