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": 76030.86519024103,
-      "required_pages_per_second_upper_estimate": 0.006590162696224003
+      "required_token_positions_per_second": 75440.0863967254,
+      "required_pages_per_second_upper_estimate": 0.006538955487717231
     },
     {
       "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": 9503.85814878013,
-      "required_pages_per_second_upper_estimate": 0.0037912311108904296
+      "required_token_positions_per_second": 9430.010799590675,
+      "required_pages_per_second_upper_estimate": 0.0037617722991824935
     },
     {
       "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": 19007.71629756026,
-      "required_pages_per_second_upper_estimate": 0.004397100641059155
+      "required_token_positions_per_second": 18860.02159918135,
+      "required_pages_per_second_upper_estimate": 0.004362934071927107
     },
     {
       "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": 28511.574446340386,
-      "required_pages_per_second_upper_estimate": 0.0017362052951118312
+      "required_token_positions_per_second": 28290.032398772026,
+      "required_pages_per_second_upper_estimate": 0.001722714546756211
     },
     {
       "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": 28511.574446340386,
-      "required_pages_per_second_upper_estimate": 0.002194899925968781
+      "required_token_positions_per_second": 28290.032398772026,
+      "required_pages_per_second_upper_estimate": 0.00217784500587933
     },
     {
       "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": 28511.574446340386,
-      "required_pages_per_second_upper_estimate": 0.35401833252219955
+      "required_token_positions_per_second": 28290.032398772026,
+      "required_pages_per_second_upper_estimate": 0.3512675217449374
     }
   ],
   "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": 0.1225492097264862,
-    "producer_p95_seconds": 0.30686535499989986,
-    "gpu_update_seconds": 0.9697114430504125,
-    "producer_batches_per_second": 8.15998734085572,
-    "gpu_batches_per_second": 1.031234608157566,
-    "producer_headroom": 7.912833099574298,
+    "producer_batch_seconds": 0.09977969827343713,
+    "producer_p95_seconds": 0.3288669440007652,
+    "gpu_update_seconds": 0.9773053494700171,
+    "producer_batches_per_second": 10.022078812661785,
+    "gpu_batches_per_second": 1.0232216579416964,
+    "producer_headroom": 9.79463123642448,
     "sustainable": true,
-    "buffer_seconds": 60.122109469125576,
+    "buffer_seconds": 60.59293166714106,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 401715.9304770413,
-    "upload_bytes_per_second": 3467499.379001222,
+    "generation_bytes_per_second": 418398.62041856704,
+    "upload_bytes_per_second": 3431159.4409802807,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.19888721211842,
+    "estimated_full_shards_per_hour": 89.77860412042388,
     "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": 0.9226963619985327,
-      "bytes": 5611919,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 0.7924021649996575,
       "bytes": 1122256,
@@ -562,6 +556,12 @@
       "bytes": 11859045,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 0.8893314249980904,
+      "bytes": 1064415,
+      "pages": 30,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -265,3 +265,4 @@
 {"step": 66500, "loss": 0.5951087810099125, "lr": 0.0003, "grad_norm": 0.13952694373766053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9732476774, "tokens_per_second": 148525.42918499935, "data_wait_s": 1.1972999345744029e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 66600, "loss": 0.5865750210359693, "lr": 0.0003, "grad_norm": 0.1314899339702683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9747201222, "tokens_per_second": 150907.16278123777, "data_wait_s": 1.5180001355474815e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 66700, "loss": 0.6159664392471313, "lr": 0.0003, "grad_norm": 0.1255530404184478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9761906396, "tokens_per_second": 149765.30137521963, "data_wait_s": 7.580001692986116e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 66800, "loss": 0.6008422300219536, "lr": 0.0003, "grad_norm": 0.1434360788227507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9776487304, "tokens_per_second": 148791.11147907007, "data_wait_s": 1.032299769576639e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7207650e15d4c3280b820bccb400c6e0e88f083937f5ac8d9690e2d06473d92b
-size 569586
+oid sha256:2011d9738fb7187404dca94ad3b56a052fa9028f8d0450a740c333f9491da4fc
+size 571704