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": 76303.90861097998,
-      "required_pages_per_second_upper_estimate": 0.005278881447168228
+      "required_token_positions_per_second": 77220.68572185696,
+      "required_pages_per_second_upper_estimate": 0.005342306214914668
     },
     {
       "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": 9537.988576372498,
-      "required_pages_per_second_upper_estimate": 0.003993356641076649
+      "required_token_positions_per_second": 9652.58571523212,
+      "required_pages_per_second_upper_estimate": 0.00404133607005679
     },
     {
       "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": 19075.977152744996,
-      "required_pages_per_second_upper_estimate": 0.003973022973569451
+      "required_token_positions_per_second": 19305.17143046424,
+      "required_pages_per_second_upper_estimate": 0.004020758097358807
     },
     {
       "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": 28613.965729117495,
-      "required_pages_per_second_upper_estimate": 0.0017153635544647162
+      "required_token_positions_per_second": 28957.757145696363,
+      "required_pages_per_second_upper_estimate": 0.001735973324949521
     },
     {
       "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": 28613.965729117495,
-      "required_pages_per_second_upper_estimate": 0.0022246099759048884
+      "required_token_positions_per_second": 28957.757145696363,
+      "required_pages_per_second_upper_estimate": 0.0022513382463652657
     },
     {
       "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": 28613.965729117495,
-      "required_pages_per_second_upper_estimate": 0.3670292290904105
+      "required_token_positions_per_second": 28957.757145696363,
+      "required_pages_per_second_upper_estimate": 0.3714390162478209
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.08879371699214289,
-    "producer_p95_seconds": 0.2115945069999725,
-    "gpu_update_seconds": 0.9662414592139345,
-    "producer_batches_per_second": 11.26205810359856,
-    "gpu_batches_per_second": 1.0349379965681964,
-    "producer_headroom": 10.881867455773188,
+    "producer_batch_seconds": 0.11025657638302278,
+    "producer_p95_seconds": 0.2322854100020777,
+    "gpu_update_seconds": 0.9547700763181856,
+    "producer_batches_per_second": 9.069753776192703,
+    "gpu_batches_per_second": 1.0473725819479298,
+    "producer_headroom": 8.65952950508266,
     "sustainable": true,
-    "buffer_seconds": 60.87321193047787,
+    "buffer_seconds": 60.15051480804569,
     "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": 347320.59519967146,
-    "upload_bytes_per_second": 4757526.033105226,
+    "generation_bytes_per_second": 347050.47687860526,
+    "upload_bytes_per_second": 4773889.065568077,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.52691452019317,
+    "estimated_full_shards_per_hour": 74.46895341652507,
     "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": 1.29576844300027,
-      "bytes": 7786718,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0071225869996852,
       "bytes": 672738,
@@ -562,6 +556,12 @@
       "bytes": 10469225,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.9355267249993631,
+      "bytes": 7522148,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -185,3 +185,4 @@
 {"step": 415600, "loss": 0.5740628838539124, "lr": 0.0003, "grad_norm": 0.14756818923425744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60832757758, "tokens_per_second": 153535.41750675783, "data_wait_s": 8.146002073772252e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 415700, "loss": 0.5602020025253296, "lr": 0.0003, "grad_norm": 0.150553616427438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60847417908, "tokens_per_second": 152569.99292592288, "data_wait_s": 1.350199818261899e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 415800, "loss": 0.6331706643104553, "lr": 0.0003, "grad_norm": 0.1568104825987768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60862157194, "tokens_per_second": 153088.32221817953, "data_wait_s": 1.1731997801689431e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 415900, "loss": 0.5758548378944397, "lr": 0.0003, "grad_norm": 0.15755216020757526, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60876652439, "tokens_per_second": 117081.74765913683, "data_wait_s": 9.520001185592264e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:790e0e000aab2ff0e3915eb69cc6a1d0d18b2c6771549a15fee6a0f28241bcc2
-size 399363
+oid sha256:338b1477f121b03f9ecd3ea6e5e901984d7b1af15256e4619ba9625a39a0b313
+size 401481