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": 76355.68696623994,
-      "required_pages_per_second_upper_estimate": 0.005282463593927448
+      "required_token_positions_per_second": 76954.13948611787,
+      "required_pages_per_second_upper_estimate": 0.005323865927853761
     },
     {
       "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": 9544.460870779993,
-      "required_pages_per_second_upper_estimate": 0.003996066455588191
+      "required_token_positions_per_second": 9619.267435764734,
+      "required_pages_per_second_upper_estimate": 0.0040273864022085885
     },
     {
       "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": 19088.921741559985,
-      "required_pages_per_second_upper_estimate": 0.00397571899004785
+      "required_token_positions_per_second": 19238.534871529468,
+      "required_pages_per_second_upper_estimate": 0.004006879459456928
     },
     {
       "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": 28633.382612339978,
-      "required_pages_per_second_upper_estimate": 0.0017165275669660401
+      "required_token_positions_per_second": 28857.8023072942,
+      "required_pages_per_second_upper_estimate": 0.0017299811850095127
     },
     {
       "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": 28633.382612339978,
-      "required_pages_per_second_upper_estimate": 0.002226119553169593
+      "required_token_positions_per_second": 28857.8023072942,
+      "required_pages_per_second_upper_estimate": 0.0022435671973343677
     },
     {
       "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": 28633.382612339978,
-      "required_pages_per_second_upper_estimate": 0.36727828802016366
+      "required_token_positions_per_second": 28857.8023072942,
+      "required_pages_per_second_upper_estimate": 0.3701569029039417
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.07693654072658518,
-    "producer_p95_seconds": 0.17599810100000468,
-    "gpu_update_seconds": 0.9655862310897452,
-    "producer_batches_per_second": 12.997725015396398,
-    "gpu_batches_per_second": 1.0356402854578985,
-    "producer_headroom": 12.550424310357508,
+    "producer_batch_seconds": 0.45304742894534655,
+    "producer_p95_seconds": 0.18793714299954445,
+    "gpu_update_seconds": 0.9580771156995415,
+    "producer_batches_per_second": 2.207274417885804,
+    "gpu_batches_per_second": 1.0437573172487775,
+    "producer_headroom": 2.1147391078454154,
     "sustainable": true,
-    "buffer_seconds": 60.831932558653946,
+    "buffer_seconds": 60.358858289071115,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 376187.947467435,
-    "upload_bytes_per_second": 2904466.2374590193,
+    "generation_bytes_per_second": 328459.3244965244,
+    "upload_bytes_per_second": 2948605.3177353926,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.72117631928718,
+    "estimated_full_shards_per_hour": 70.47972489520835,
     "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": [
@@ -400,6 +400,12 @@
       "bytes": 7690441,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 0.9298561730001893,
+      "bytes": 6431657,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -78,3 +78,4 @@
 {"step": 652800, "loss": 0.5506278872489929, "lr": 0.0001, "grad_norm": 0.21742603238567818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95552335978, "tokens_per_second": 152711.57912437886, "data_wait_s": 1.398899985360913e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 652900, "loss": 0.5208961963653564, "lr": 0.0001, "grad_norm": 0.17214884091978144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95567013116, "tokens_per_second": 153149.21776560624, "data_wait_s": 1.3562000276579056e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 653000, "loss": 0.5420287251472473, "lr": 0.0001, "grad_norm": 0.18063351481286485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95581694897, "tokens_per_second": 153296.25834348908, "data_wait_s": 7.841999831725843e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 653100, "loss": 0.6101648807525635, "lr": 0.0001, "grad_norm": 0.20305761003838585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95596603474, "tokens_per_second": 156076.9113091464, "data_wait_s": 9.385001249029301e-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:de6c101b265d99622ce7fd2dddb2f24c73aaae538b885c124b7d0d77a0e30536
-size 170388
+oid sha256:5a7e97899c8c449b32e955bf34f0ae175637496893de2f4986ec1e6db5ecba09
+size 172506