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": 77171.55873625647,
-      "required_pages_per_second_upper_estimate": 0.005338907496060514
+      "required_token_positions_per_second": 77201.67480760421,
+      "required_pages_per_second_upper_estimate": 0.005340990995755264
     },
     {
       "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": 9646.444842032059,
-      "required_pages_per_second_upper_estimate": 0.004038765014683939
+      "required_token_positions_per_second": 9650.209350950527,
+      "required_pages_per_second_upper_estimate": 0.004040341135956215
     },
     {
       "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": 19292.889684064117,
-      "required_pages_per_second_upper_estimate": 0.004018200133475095
+      "required_token_positions_per_second": 19300.418701901053,
+      "required_pages_per_second_upper_estimate": 0.0040197682293369764
     },
     {
       "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": 28939.334526096176,
-      "required_pages_per_second_upper_estimate": 0.0017348689170342013
+      "required_token_positions_per_second": 28950.62805285158,
+      "required_pages_per_second_upper_estimate": 0.001735545946718971
     },
     {
       "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": 28939.334526096176,
-      "required_pages_per_second_upper_estimate": 0.002249905968723891
+      "required_token_positions_per_second": 28950.62805285158,
+      "required_pages_per_second_upper_estimate": 0.0022507839907541494
     },
     {
       "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": 28939.334526096176,
-      "required_pages_per_second_upper_estimate": 0.3712027106642575
+      "required_token_positions_per_second": 28950.62805285158,
+      "required_pages_per_second_upper_estimate": 0.3713475718994315
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.25640987414033134,
-    "producer_p95_seconds": 0.33060542300518136,
-    "gpu_update_seconds": 0.9553778776449849,
-    "producer_batches_per_second": 3.9000058143342287,
-    "gpu_batches_per_second": 1.0467062545607702,
-    "producer_headroom": 3.7259792777017364,
+    "producer_batch_seconds": 0.11203869885980566,
+    "producer_p95_seconds": 0.336805702005222,
+    "gpu_update_seconds": 0.955005188472128,
+    "producer_batches_per_second": 8.925487444756055,
+    "gpu_batches_per_second": 1.0471147299208472,
+    "producer_headroom": 8.523886819384868,
     "sustainable": true,
-    "buffer_seconds": 60.18880629163405,
+    "buffer_seconds": 60.165326873744064,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 451844.8356642341,
-    "upload_bytes_per_second": 4800530.976203412,
+    "generation_bytes_per_second": 452710.60318855784,
+    "upload_bytes_per_second": 4746473.526917738,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.95538332410113,
+    "estimated_full_shards_per_hour": 97.14115688078452,
     "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": 1.268981666999025,
-      "bytes": 11497273,
-      "pages": 33,
-      "shards": 4
-    },
     {
       "seconds": 1.0184783190015878,
       "bytes": 1341723,
@@ -562,6 +556,12 @@
       "bytes": 7228890,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.1513834180004778,
+      "bytes": 6077960,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -346,3 +346,4 @@
 {"step": 959600, "loss": 0.5301459431648254, "lr": 3e-05, "grad_norm": 0.344852297676844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140459388377, "tokens_per_second": 150807.61525164032, "data_wait_s": 1.1290001566521823e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 959700, "loss": 0.4841211438179016, "lr": 3e-05, "grad_norm": 0.34537294225829357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140473890672, "tokens_per_second": 150508.60422501882, "data_wait_s": 1.7505000869277865e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 959800, "loss": 0.4464470148086548, "lr": 3e-05, "grad_norm": 0.4452609082534058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140488578085, "tokens_per_second": 153033.0980581865, "data_wait_s": 1.3746001059189439e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 959900, "loss": 0.6030485033988953, "lr": 3e-05, "grad_norm": 0.3866498325193008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140503235806, "tokens_per_second": 152363.01700591584, "data_wait_s": 1.022600190481171e-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 sha256:3db77757a101a008c9e5ca0e015eb04c273e1bf7c866c0039a823cdba774fb69
-size 741927
+oid sha256:d9518617f46b8172e640ea4ab3afe9d965b470fad1793bbda92275e30a6e79ba
+size 744045