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": 76910.40217973563,
-      "required_pages_per_second_upper_estimate": 0.005320840079513706
+      "required_token_positions_per_second": 77026.8316224426,
+      "required_pages_per_second_upper_estimate": 0.005328894938513701
     },
     {
       "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": 9613.800272466953,
-      "required_pages_per_second_upper_estimate": 0.004025097415103159
+      "required_token_positions_per_second": 9628.353952805324,
+      "required_pages_per_second_upper_estimate": 0.004031190737145431
     },
     {
       "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": 19227.600544933906,
-      "required_pages_per_second_upper_estimate": 0.004004602127584656
+      "required_token_positions_per_second": 19256.70790561065,
+      "required_pages_per_second_upper_estimate": 0.004010664423200901
     },
     {
       "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": 28841.40081740086,
-      "required_pages_per_second_upper_estimate": 0.0017289979407340318
+      "required_token_positions_per_second": 28885.06185841597,
+      "required_pages_per_second_upper_estimate": 0.0017316153534763382
     },
     {
       "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": 28841.40081740086,
-      "required_pages_per_second_upper_estimate": 0.00224229205363769
+      "required_token_positions_per_second": 28885.06185841597,
+      "required_pages_per_second_upper_estimate": 0.0022456865075319915
     },
     {
       "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": 28841.40081740086,
-      "required_pages_per_second_upper_estimate": 0.36994652220213775
+      "required_token_positions_per_second": 28885.06185841597,
+      "required_pages_per_second_upper_estimate": 0.37050655915670616
     }
   ],
   "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": 0.2554917378203072,
-    "producer_p95_seconds": 0.3272475379999378,
-    "gpu_update_seconds": 0.9586219537339238,
-    "producer_batches_per_second": 3.9140208937140715,
-    "gpu_batches_per_second": 1.0431640920645566,
-    "producer_headroom": 3.7520663560875813,
+    "producer_batch_seconds": 0.08628433124215462,
+    "producer_p95_seconds": 0.3091390580011648,
+    "gpu_update_seconds": 0.9571729545022407,
+    "producer_batches_per_second": 11.589589738993599,
+    "gpu_batches_per_second": 1.0447432674484944,
+    "producer_headroom": 11.093241851941356,
     "sustainable": true,
-    "buffer_seconds": 60.3931830852372,
+    "buffer_seconds": 60.301896133641165,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338805.1872725868,
+    "generation_bytes_per_second": 373144.2344595316,
     "upload_bytes_per_second": 4928704.762808889,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.69970620759204,
+    "estimated_full_shards_per_hour": 80.06806636180364,
     "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

@@ -304,3 +304,4 @@
 {"step": 955400, "loss": 0.49536147713661194, "lr": 3e-05, "grad_norm": 0.3325124770746695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139844004871, "tokens_per_second": 151892.91272191438, "data_wait_s": 1.5466997865587473e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 955500, "loss": 0.5703268647193909, "lr": 3e-05, "grad_norm": 0.35246085092226154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139858679024, "tokens_per_second": 152760.00639174634, "data_wait_s": 1.1490003089420497e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 955600, "loss": 0.5585286021232605, "lr": 3e-05, "grad_norm": 0.3537998040043666, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139873419564, "tokens_per_second": 153427.827137965, "data_wait_s": 9.121999028138816e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 955700, "loss": 0.5493685603141785, "lr": 3e-05, "grad_norm": 0.3441077539652908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139888187009, "tokens_per_second": 153591.7385881006, "data_wait_s": 8.060997060965747e-06, "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 sha256:1772df6e2932a46c09b4cf061c40867baa4ad9dcce36b95ac4f2eec9d0d3e828
-size 652971
+oid sha256:2d9a702760283555127aaeda0e9e409f95b804413eafa8e9d11452bdd40a7b01
+size 655089