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": 77207.94551668425,
-      "required_pages_per_second_upper_estimate": 0.005341424817959471
+      "required_token_positions_per_second": 77056.46564291079,
+      "required_pages_per_second_upper_estimate": 0.0053309450888100905
     },
     {
       "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": 9650.993189585532,
-      "required_pages_per_second_upper_estimate": 0.004040669312824281
+      "required_token_positions_per_second": 9632.058205363848,
+      "required_pages_per_second_upper_estimate": 0.004032741630338095
     },
     {
       "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": 19301.986379171063,
-      "required_pages_per_second_upper_estimate": 0.0040200947351698895
+      "required_token_positions_per_second": 19264.116410727696,
+      "required_pages_per_second_upper_estimate": 0.004012207419441364
     },
     {
       "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": 28952.979568756597,
-      "required_pages_per_second_upper_estimate": 0.0017356869165069198
+      "required_token_positions_per_second": 28896.174616091543,
+      "required_pages_per_second_upper_estimate": 0.0017322815463307903
     },
     {
       "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": 28952.979568756597,
-      "required_pages_per_second_upper_estimate": 0.002250966810772539
+      "required_token_positions_per_second": 28896.174616091543,
+      "required_pages_per_second_upper_estimate": 0.0022465504755614122
     },
     {
       "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": 28952.979568756597,
-      "required_pages_per_second_upper_estimate": 0.3713777346206
+      "required_token_positions_per_second": 28896.174616091543,
+      "required_pages_per_second_upper_estimate": 0.37064910168021886
     }
   ],
   "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": 0.36766646025014893,
-    "producer_p95_seconds": 0.21549671399952786,
-    "gpu_update_seconds": 0.9549276244381836,
-    "producer_batches_per_second": 2.7198564680597483,
-    "gpu_batches_per_second": 1.047199781856069,
-    "producer_headroom": 2.597266075857124,
+    "producer_batch_seconds": 0.13456432946883012,
+    "producer_p95_seconds": 0.22000821100118628,
+    "gpu_update_seconds": 0.9568048493382073,
+    "producer_batches_per_second": 7.4313899080635295,
+    "gpu_batches_per_second": 1.0451452045750702,
+    "producer_headroom": 7.110389901358199,
     "sustainable": true,
-    "buffer_seconds": 60.160440339605564,
+    "buffer_seconds": 60.27870550830706,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364588.9727559351,
+    "generation_bytes_per_second": 362456.83696716215,
     "upload_bytes_per_second": 3659985.693033039,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.2323063565115,
+    "estimated_full_shards_per_hour": 77.77479964982174,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -163,3 +163,4 @@
 {"step": 941300, "loss": 0.500248908996582, "lr": 3e-05, "grad_norm": 0.3266684065896893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137779724167, "tokens_per_second": 153474.19305677293, "data_wait_s": 1.4176999684423208e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 941400, "loss": 0.5718052387237549, "lr": 3e-05, "grad_norm": 0.3593922088571367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137794430820, "tokens_per_second": 147108.89537589918, "data_wait_s": 1.6684000001987442e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 941500, "loss": 0.4808211326599121, "lr": 3e-05, "grad_norm": 0.3368848088417733, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137808972846, "tokens_per_second": 151134.82237175028, "data_wait_s": 1.4888999430695549e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 941600, "loss": 0.5177056789398193, "lr": 3e-05, "grad_norm": 0.3529110804909721, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137823537238, "tokens_per_second": 152451.43411521966, "data_wait_s": 1.446100213797763e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7a5cd4e72aa5a90e5c57487d6c7fc0bcab65fbd89dbbe50f2889ae67357f1b9f
-size 351984
+oid sha256:72fa1efda29f7c178d01541fe8adbd3e83d9589fcea1502385c2cfe07d9963ef
+size 354102