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": 76458.91533972324,
-      "required_pages_per_second_upper_estimate": 0.00528960517232785
+      "required_token_positions_per_second": 77312.23164535525,
+      "required_pages_per_second_upper_estimate": 0.005348639574318087
     },
     {
       "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": 9557.364417465405,
-      "required_pages_per_second_upper_estimate": 0.004001468901128677
+      "required_token_positions_per_second": 9664.028955669406,
+      "required_pages_per_second_upper_estimate": 0.004046127115865847
     },
     {
       "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": 19114.72883493081,
-      "required_pages_per_second_upper_estimate": 0.003981093927018174
+      "required_token_positions_per_second": 19328.057911338812,
+      "required_pages_per_second_upper_estimate": 0.004025524747767893
     },
     {
       "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": 28672.09325239622,
-      "required_pages_per_second_upper_estimate": 0.0017188482107296868
+      "required_token_positions_per_second": 28992.086867008216,
+      "required_pages_per_second_upper_estimate": 0.0017380313393237175
     },
     {
       "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": 28672.09325239622,
-      "required_pages_per_second_upper_estimate": 0.0022291291351638548
+      "required_token_positions_per_second": 28992.086867008216,
+      "required_pages_per_second_upper_estimate": 0.0022540072311967767
     },
     {
       "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": 28672.09325239622,
-      "required_pages_per_second_upper_estimate": 0.36777482654655813
+      "required_token_positions_per_second": 28992.086867008216,
+      "required_pages_per_second_upper_estimate": 0.3718793610524264
     }
   ],
   "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": 0.11594154529689149,
-    "producer_p95_seconds": 0.3170849010020902,
-    "gpu_update_seconds": 0.9642825780670676,
-    "producer_batches_per_second": 8.625035982048542,
-    "gpu_batches_per_second": 1.037040409881229,
-    "producer_headroom": 8.31697193269099,
+    "producer_batch_seconds": 0.11475533454682818,
+    "producer_p95_seconds": 0.30705034299899125,
+    "gpu_update_seconds": 0.953639526772468,
+    "producer_batches_per_second": 8.714191840832726,
+    "gpu_batches_per_second": 1.0486142530023226,
+    "producer_headroom": 8.310197783296223,
     "sustainable": true,
-    "buffer_seconds": 60.74980241822526,
+    "buffer_seconds": 60.07929018666548,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 347634.5000443587,
-    "upload_bytes_per_second": 5657364.624840701,
+    "generation_bytes_per_second": 347263.73793835,
+    "upload_bytes_per_second": 5642097.823338794,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.59427119253226,
+    "estimated_full_shards_per_hour": 74.51471427548289,
     "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": 1.1533994220008026,
-      "bytes": 7669461,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.8288239209978201,
       "bytes": 961455,
@@ -562,6 +556,12 @@
       "bytes": 2487942,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.1110187749982288,
+      "bytes": 6282739,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -320,3 +320,4 @@
 {"step": 429100, "loss": 0.6631481051445007, "lr": 0.0003, "grad_norm": 0.17248152837054945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62808446446, "tokens_per_second": 150982.356583782, "data_wait_s": 2.0313000277383253e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 429200, "loss": 0.5594878196716309, "lr": 0.0003, "grad_norm": 0.13760300670004028, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62822966979, "tokens_per_second": 150400.38467674743, "data_wait_s": 1.567399885971099e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 429300, "loss": 0.5749633312225342, "lr": 0.0003, "grad_norm": 0.1391939607394664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62837539403, "tokens_per_second": 151868.3327737305, "data_wait_s": 1.2073996913386509e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 429400, "loss": 0.5377009510993958, "lr": 0.0003, "grad_norm": 0.14921407700600425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62852273036, "tokens_per_second": 152181.0550588608, "data_wait_s": 1.119099761126563e-05, "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 sha256:046fd15efe85edd5f4291548ae8df28b3618fc0747abcff75b86fe08bd4e2710
-size 686859
+oid sha256:29109f26199e2c8852e232c3e911f97ebf6515283b86478e462ba69e547dc458
+size 688977