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": 75506.21521538874,
-      "required_pages_per_second_upper_estimate": 0.006544687366117011
+      "required_token_positions_per_second": 75457.98407937646,
+      "required_pages_per_second_upper_estimate": 0.006540506813488163
     },
     {
       "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": 9438.276901923593,
-      "required_pages_per_second_upper_estimate": 0.003765069770992338
+      "required_token_positions_per_second": 9432.248009922057,
+      "required_pages_per_second_upper_estimate": 0.003762664755832957
     },
     {
       "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": 18876.553803847186,
-      "required_pages_per_second_upper_estimate": 0.004366758506519684
+      "required_token_positions_per_second": 18864.496019844115,
+      "required_pages_per_second_upper_estimate": 0.004363969150400324
     },
     {
       "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": 28314.83070577078,
-      "required_pages_per_second_upper_estimate": 0.0017242246335457735
+      "required_token_positions_per_second": 28296.744029766174,
+      "required_pages_per_second_upper_estimate": 0.0017231232498705483
     },
     {
       "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": 28314.83070577078,
-      "required_pages_per_second_upper_estimate": 0.0021797540481982014
+      "required_token_positions_per_second": 28296.744029766174,
+      "required_pages_per_second_upper_estimate": 0.002178361685812243
     },
     {
       "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": 28314.83070577078,
-      "required_pages_per_second_upper_estimate": 0.3515754337232673
+      "required_token_positions_per_second": 28296.744029766174,
+      "required_pages_per_second_upper_estimate": 0.35135085773950075
     }
   ],
   "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": 0.08376432873430417,
-    "producer_p95_seconds": 0.3388251029973617,
-    "gpu_update_seconds": 0.9764494192919587,
-    "producer_batches_per_second": 11.938255998826717,
-    "gpu_batches_per_second": 1.0241185874483065,
-    "producer_headroom": 11.657103137413092,
+    "producer_batch_seconds": 0.15747545338282976,
+    "producer_p95_seconds": 0.36809645099856425,
+    "gpu_update_seconds": 0.9770735449603764,
+    "producer_batches_per_second": 6.35019603702271,
+    "gpu_batches_per_second": 1.0234644107988344,
+    "producer_headroom": 6.2046085530871125,
     "sustainable": true,
-    "buffer_seconds": 60.53986399610144,
+    "buffer_seconds": 60.578559787543334,
     "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": 317188.75501410925,
-    "upload_bytes_per_second": 6320216.913289845,
+    "generation_bytes_per_second": 326674.22975049843,
+    "upload_bytes_per_second": 6448521.905457721,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.06132304971179,
+    "estimated_full_shards_per_hour": 70.09668511759008,
     "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": 1.020132988000114,
-      "bytes": 882161,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.030573132000427,
       "bytes": 11789498,
@@ -562,6 +556,12 @@
       "bytes": 7632407,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.273418650001986,
+      "bytes": 11058609,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -307,3 +307,4 @@
 {"step": 30700, "loss": 0.637079119682312, "lr": 0.0003, "grad_norm": 0.12613209870492187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4492616372, "tokens_per_second": 148820.0524365116, "data_wait_s": 1.049999991664663e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 30800, "loss": 0.6641022209078074, "lr": 0.0003, "grad_norm": 0.1305812118695991, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4507160282, "tokens_per_second": 147366.82299839662, "data_wait_s": 1.0823998309206218e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 30900, "loss": 0.6125746350735426, "lr": 0.0003, "grad_norm": 0.128221784809622, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4521759699, "tokens_per_second": 149466.23070962646, "data_wait_s": 1.0699000995373353e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 31000, "loss": 0.6418987549841404, "lr": 0.0003, "grad_norm": 0.1280485221356012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4536310497, "tokens_per_second": 147922.7351113262, "data_wait_s": 1.1928001185879111e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fea7ae3f47a651edacfcc006c46521aded39021db3a39d5e50ae1f1da3158c08
-size 653551
+oid sha256:cc6bb9128e069ee029b53474bf8ace15b91ee599a796c54e77b45552f90b3f87
+size 655669