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": 76741.42228590837,
-      "required_pages_per_second_upper_estimate": 0.005309149658371363
+      "required_token_positions_per_second": 77577.90939393648,
+      "required_pages_per_second_upper_estimate": 0.005367019777422244
     },
     {
       "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": 9592.677785738546,
-      "required_pages_per_second_upper_estimate": 0.004016253870997656
+      "required_token_positions_per_second": 9697.23867424206,
+      "required_pages_per_second_upper_estimate": 0.004060031331534434
     },
     {
       "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": 19185.355571477092,
-      "required_pages_per_second_upper_estimate": 0.00399580361368847
+      "required_token_positions_per_second": 19394.47734848412,
+      "required_pages_per_second_upper_estimate": 0.004039358164926913
     },
     {
       "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": 28778.033357215638,
-      "required_pages_per_second_upper_estimate": 0.0017251991582524372
+      "required_token_positions_per_second": 29091.71602272618,
+      "required_pages_per_second_upper_estimate": 0.001744003955084097
     },
     {
       "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": 28778.033357215638,
-      "required_pages_per_second_upper_estimate": 0.00223736551233229
+      "required_token_positions_per_second": 29091.71602272618,
+      "required_pages_per_second_upper_estimate": 0.002261752959831505
     },
     {
       "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": 28778.033357215638,
-      "required_pages_per_second_upper_estimate": 0.36913371246155946
+      "required_token_positions_per_second": 29091.71602272618,
+      "required_pages_per_second_upper_estimate": 0.3731572968885235
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11029548517973353,
-    "producer_p95_seconds": 0.32175242199446075,
-    "gpu_update_seconds": 0.9607327803401722,
-    "producer_batches_per_second": 9.066554250796722,
-    "gpu_batches_per_second": 1.0408721555705889,
-    "producer_headroom": 8.710535873472942,
+    "producer_batch_seconds": 0.1407548160155443,
+    "producer_p95_seconds": 0.332064309004636,
+    "gpu_update_seconds": 0.9503736382687648,
+    "producer_batches_per_second": 7.104552642018052,
+    "gpu_batches_per_second": 1.0522177380905013,
+    "producer_headroom": 6.751979542666661,
     "sustainable": true,
-    "buffer_seconds": 60.526165161430846,
+    "buffer_seconds": 60.823912849200944,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 458201.42047916865,
-    "upload_bytes_per_second": 5060465.756765956,
+    "generation_bytes_per_second": 457494.1428036806,
+    "upload_bytes_per_second": 5046843.635732153,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.31935845166487,
+    "estimated_full_shards_per_hour": 98.16759312708677,
     "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": 1.1813513000015519,
-      "bytes": 7573359,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 1.4761009860012564,
       "bytes": 5975966,
@@ -562,6 +556,12 @@
       "bytes": 11305199,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.3714339009966352,
+      "bytes": 7352671,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -380,3 +380,4 @@
 {"step": 683000, "loss": 0.5247899889945984, "lr": 0.0001, "grad_norm": 0.20268397962436385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99973492914, "tokens_per_second": 151112.64794044953, "data_wait_s": 1.226499443873763e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 683100, "loss": 0.540785551071167, "lr": 0.0001, "grad_norm": 0.2015798868947121, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99988027993, "tokens_per_second": 151482.1991421006, "data_wait_s": 8.834998880047351e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 683200, "loss": 0.5725772380828857, "lr": 0.0001, "grad_norm": 0.2014802505878475, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100002652287, "tokens_per_second": 153065.17170915345, "data_wait_s": 1.7001002561300993e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 683300, "loss": 0.5685223937034607, "lr": 0.0001, "grad_norm": 0.20739122870950957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100017359825, "tokens_per_second": 152272.69776061346, "data_wait_s": 1.486000110162422e-05, "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 sha256:09c8d988fb629290f165c5821f9eab1b1f675cbdcf7dd29cf202c9cb3fc17143
-size 814722
+oid sha256:0325e10f4bd210cbc736a4dcadcd65710c3a2e06f82d68da27fb1294dc667093
+size 816840