Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75265.90373660412,
-      "required_pages_per_second_upper_estimate": 0.0065238577761998345
+      "required_token_positions_per_second": 75047.24486470953,
+      "required_pages_per_second_upper_estimate": 0.006504904979369856
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9408.237967075514,
-      "required_pages_per_second_upper_estimate": 0.0037530867907593405
+      "required_token_positions_per_second": 9380.905608088691,
+      "required_pages_per_second_upper_estimate": 0.0037421835041043126
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18816.47593415103,
-      "required_pages_per_second_upper_estimate": 0.004352860548699872
+      "required_token_positions_per_second": 18761.811216177382,
+      "required_pages_per_second_upper_estimate": 0.0043402148282628466
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28224.713901226543,
-      "required_pages_per_second_upper_estimate": 0.0017187369929553627
+      "required_token_positions_per_second": 28142.716824266074,
+      "required_pages_per_second_upper_estimate": 0.0017137438011738848
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28224.713901226543,
-      "required_pages_per_second_upper_estimate": 0.002172816607125105
+      "required_token_positions_per_second": 28142.716824266074,
+      "required_pages_per_second_upper_estimate": 0.0021665042451582514
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28224.713901226543,
-      "required_pages_per_second_upper_estimate": 0.3504564846123713
+      "required_token_positions_per_second": 28142.716824266074,
+      "required_pages_per_second_upper_estimate": 0.3494383553430855
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2776922882655981,
-    "producer_p95_seconds": 0.2755860910001502,
-    "gpu_update_seconds": 0.9795670594485112,
-    "producer_batches_per_second": 3.601108285166178,
-    "gpu_batches_per_second": 1.0208591544135757,
-    "producer_headroom": 3.527527053655904,
+    "producer_batch_seconds": 0.0953098191562276,
+    "producer_p95_seconds": 0.32717703399976017,
+    "gpu_update_seconds": 0.9824211419474788,
+    "producer_batches_per_second": 10.49209838873836,
+    "gpu_batches_per_second": 1.017893403655457,
+    "producer_headroom": 10.30765928048964,
     "sustainable": true,
-    "buffer_seconds": 60.7331576858077,
+    "buffer_seconds": 60.91011080074368,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 417183.6655558831,
-    "upload_bytes_per_second": 3976100.285937116,
+    "generation_bytes_per_second": 417334.3071955894,
+    "upload_bytes_per_second": 3917774.146267601,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.51790308959359,
+    "estimated_full_shards_per_hour": 89.55022727871668,
     "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": 1.0978697010014002,
-      "bytes": 12068565,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 0.9089816479990986,
       "bytes": 779851,
@@ -562,6 +556,12 @@
       "bytes": 7251816,
       "pages": 27,
       "shards": 3
+    },
+    {
+      "seconds": 1.169119137000962,
+      "bytes": 6109087,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -226,3 +226,4 @@
 {"step": 137600, "loss": 0.5732929743826389, "lr": 0.0003, "grad_norm": 0.1366899343635189, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20137266082, "tokens_per_second": 149694.05483137074, "data_wait_s": 1.8479000573279336e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 137700, "loss": 0.5929130781441927, "lr": 0.0003, "grad_norm": 0.13340303257376512, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20151980362, "tokens_per_second": 150106.5723258953, "data_wait_s": 1.0334999387850985e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 137800, "loss": 0.599891185760498, "lr": 0.0003, "grad_norm": 0.13112477463219238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20166656510, "tokens_per_second": 149432.94795432422, "data_wait_s": 1.263399826711975e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 137900, "loss": 0.661179369315505, "lr": 0.0003, "grad_norm": 0.15528800027034226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20181428733, "tokens_per_second": 150458.80736417428, "data_wait_s": 1.4265002391766757e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:66ff484e27584f89c83c0e3da19e93fcb32b4d20103c901b37ac0ced44f879ac
-size 486201
+oid sha256:c6ff9f10064fbe62e853e0cbaed0eef5db2dbedab0c27933512202635d7418d8
+size 488319