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": 78114.71930383479,
-      "required_pages_per_second_upper_estimate": 0.005404157532559607
+      "required_token_positions_per_second": 77691.32060752469,
+      "required_pages_per_second_upper_estimate": 0.005374865828328548
     },
     {
       "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": 9764.339912979349,
-      "required_pages_per_second_upper_estimate": 0.004088125218960533
+      "required_token_positions_per_second": 9711.415075940586,
+      "required_pages_per_second_upper_estimate": 0.004065966694888676
     },
     {
       "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": 19528.679825958698,
-      "required_pages_per_second_upper_estimate": 0.004067309001827538
+      "required_token_positions_per_second": 19422.83015188117,
+      "required_pages_per_second_upper_estimate": 0.0040452633061608114
     },
     {
       "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": 29293.019738938045,
-      "required_pages_per_second_upper_estimate": 0.001756071805498021
+      "required_token_positions_per_second": 29134.245227821757,
+      "required_pages_per_second_upper_estimate": 0.0017465535159912407
     },
     {
       "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": 29293.019738938045,
-      "required_pages_per_second_upper_estimate": 0.0022774034383255064
+      "required_token_positions_per_second": 29134.245227821757,
+      "required_pages_per_second_upper_estimate": 0.0022650594184614827
     },
     {
       "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": 29293.019738938045,
-      "required_pages_per_second_upper_estimate": 0.3757394048169988
+      "required_token_positions_per_second": 29134.245227821757,
+      "required_pages_per_second_upper_estimate": 0.3737028158671869
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,27 +158,34 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 1.120676024002023,
-    "producer_p95_seconds": 2.0243115949997446,
-    "gpu_update_seconds": 0.9438426029955735,
-    "producer_batches_per_second": 0.8923185457550172,
-    "gpu_batches_per_second": 1.0594986884743216,
-    "producer_headroom": 0.8422082589266403,
-    "sustainable": false,
-    "buffer_seconds": 60.40592659171671,
-    "buffer_drain_seconds": 382.82058478354134,
+    "producer_batch_seconds": 0.0755294930470427,
+    "producer_p95_seconds": 0.2779548530015745,
+    "gpu_update_seconds": 0.9489863143458933,
+    "producer_batches_per_second": 13.23986114109307,
+    "gpu_batches_per_second": 1.0537559761220254,
+    "producer_headroom": 12.564447026737326,
+    "sustainable": true,
+    "buffer_seconds": 60.73512411813717,
+    "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 7,
-    "max_pending_pages": 1737,
-    "generation_bytes_per_second": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "upload_pages_per_commit": 22,
+    "max_pending_pages": 4096,
+    "generation_bytes_per_second": 335223.0274232138,
+    "upload_bytes_per_second": 64226.71445014656,
+    "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 71.93105809233008,
     "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": [],
+  "upload_measurements": [
+    {
+      "seconds": 2.6183652930048993,
+      "bytes": 168169,
+      "pages": 5,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

metrics/void-byte-v1/1790867478897646808-aeb36a8a/train.jsonl

@@ -0,0 +1 @@
+{"step": 440300, "loss": 0.6032773852348328, "lr": 0.0003, "grad_norm": 0.16604262242461107, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64447028234, "tokens_per_second": 151840.71188538996, "data_wait_s": 2.1765998098999262e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790867478.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.54613.0.1790867478897646808-aeb36a8a

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5b198fba435bf32e0e27d42c90fe9dad075beff7c81c3c43677cbdf4a96b8e15
-size 165
+oid sha256:6f325707ad574ba3bf31663375821e4e48061269474f77b0e09bff042f30b13d
+size 2283