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": 75977.2333082125,
-      "required_pages_per_second_upper_estimate": 0.00525628129173164
+      "required_token_positions_per_second": 75852.85018081017,
+      "required_pages_per_second_upper_estimate": 0.0052476761783693295
     },
     {
       "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": 9497.154163526562,
-      "required_pages_per_second_upper_estimate": 0.003976260125137562
+      "required_token_positions_per_second": 9481.606272601271,
+      "required_pages_per_second_upper_estimate": 0.003969750547884026
     },
     {
       "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": 18994.308327053124,
-      "required_pages_per_second_upper_estimate": 0.003956013510929601
+      "required_token_positions_per_second": 18963.212545202543,
+      "required_pages_per_second_upper_estimate": 0.003949537079621038
     },
     {
       "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": 28491.462490579688,
-      "required_pages_per_second_upper_estimate": 0.0017080196723659893
+      "required_token_positions_per_second": 28444.818817803814,
+      "required_pages_per_second_upper_estimate": 0.0017052234554038401
     },
     {
       "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": 28491.462490579688,
-      "required_pages_per_second_upper_estimate": 0.0022150858879433745
+      "required_token_positions_per_second": 28444.818817803814,
+      "required_pages_per_second_upper_estimate": 0.0022114595475488845
     },
     {
       "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": 28491.462490579688,
-      "required_pages_per_second_upper_estimate": 0.3654578890801771
+      "required_token_positions_per_second": 28444.818817803814,
+      "required_pages_per_second_upper_estimate": 0.36485959412788205
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.37978302710138223,
-    "producer_p95_seconds": 0.34908957299921894,
-    "gpu_update_seconds": 0.9703959566533811,
-    "producer_batches_per_second": 2.633082388205443,
-    "gpu_batches_per_second": 1.030507179202101,
-    "producer_headroom": 2.5551325030497902,
+    "producer_batch_seconds": 0.37294288367962736,
+    "producer_p95_seconds": 0.3932641279970994,
+    "gpu_update_seconds": 0.9719872071287344,
+    "producer_batches_per_second": 2.6813757381117895,
+    "gpu_batches_per_second": 1.0288201250652422,
+    "producer_headroom": 2.606262914950027,
     "sustainable": true,
-    "buffer_seconds": 60.16454931250963,
+    "buffer_seconds": 60.263206841981535,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354464.0869249416,
+    "generation_bytes_per_second": 334877.9224662774,
     "upload_bytes_per_second": 4921952.269350816,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.05974155245958,
+    "estimated_full_shards_per_hour": 71.85700660220375,
     "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": [

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

@@ -408,3 +408,4 @@
 {"step": 685800, "loss": 0.5752155184745789, "lr": 0.0001, "grad_norm": 0.22514589126917264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100383214867, "tokens_per_second": 152487.6991032466, "data_wait_s": 1.1365002137608826e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 685900, "loss": 0.513319730758667, "lr": 0.0001, "grad_norm": 0.19845639430238482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100397882630, "tokens_per_second": 146969.60658545097, "data_wait_s": 2.0419996872078627e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 686000, "loss": 0.5922773480415344, "lr": 0.0001, "grad_norm": 0.2260045671226013, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100412502302, "tokens_per_second": 152276.9905235283, "data_wait_s": 1.1070995242334902e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 686100, "loss": 0.59559565782547, "lr": 0.0001, "grad_norm": 0.20903594979744555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100427191750, "tokens_per_second": 152045.6171866706, "data_wait_s": 1.3903998478781432e-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:f667ba98a22216d23a0367af5751b1a6ac6494c9bfc44b4a6c68705f01b1a616
-size 874809
+oid sha256:532fda3d0908eb7e4eff8ec3eacc1018b1ce9c35810590e8f10f007663423cef
+size 876927