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": 77321.25668454487,
-      "required_pages_per_second_upper_estimate": 0.005349263947470201
+      "required_token_positions_per_second": 77341.28166839482,
+      "required_pages_per_second_upper_estimate": 0.005350649322317305
     },
     {
       "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": 9665.15708556811,
-      "required_pages_per_second_upper_estimate": 0.004046599440296403
+      "required_token_positions_per_second": 9667.660208549352,
+      "required_pages_per_second_upper_estimate": 0.004047647445617498
     },
     {
       "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": 19330.31417113622,
-      "required_pages_per_second_upper_estimate": 0.00402599466718212
+      "required_token_positions_per_second": 19335.320417098705,
+      "required_pages_per_second_upper_estimate": 0.0040270373361925215
     },
     {
       "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": 28995.47125670433,
-      "required_pages_per_second_upper_estimate": 0.0017382342283183358
+      "required_token_positions_per_second": 29002.98062564806,
+      "required_pages_per_second_upper_estimate": 0.0017386844035203698
     },
     {
       "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": 28995.47125670433,
-      "required_pages_per_second_upper_estimate": 0.0022542703526092925
+      "required_token_positions_per_second": 29002.98062564806,
+      "required_pages_per_second_upper_estimate": 0.002254854173014444
     },
     {
       "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": 28995.47125670433,
-      "required_pages_per_second_upper_estimate": 0.37192277236957366
+      "required_token_positions_per_second": 29002.98062564806,
+      "required_pages_per_second_upper_estimate": 0.3720190944914516
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.22030885208585005,
-    "producer_p95_seconds": 0.23333278000063729,
-    "gpu_update_seconds": 0.9535282167075395,
-    "producer_batches_per_second": 4.53908224990578,
-    "gpu_batches_per_second": 1.0487366629305674,
-    "producer_headroom": 4.328143003241505,
+    "producer_batch_seconds": 0.09527713949219674,
+    "producer_p95_seconds": 0.2191641060016991,
+    "gpu_update_seconds": 0.9532813319038729,
+    "producer_batches_per_second": 10.495697135007928,
+    "gpu_batches_per_second": 1.0490082691568308,
+    "producer_headroom": 10.00535214412002,
     "sustainable": true,
-    "buffer_seconds": 60.07227765257499,
+    "buffer_seconds": 60.05672390994399,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 402776.3536837642,
-    "upload_bytes_per_second": 5704718.614777752,
+    "generation_bytes_per_second": 402564.9410295652,
+    "upload_bytes_per_second": 5735685.118812759,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.4264293468923,
+    "estimated_full_shards_per_hour": 86.38106511273591,
     "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": 1.2232918650006468,
-      "bytes": 6107930,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1886329499993735,
       "bytes": 6263136,
@@ -562,6 +556,12 @@
       "bytes": 5920541,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.2547966029997042,
+      "bytes": 8617085,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -197,3 +197,4 @@
 {"step": 704700, "loss": 0.554182231426239, "lr": 8e-05, "grad_norm": 0.21033279899822088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103148434540, "tokens_per_second": 152963.37865608258, "data_wait_s": 1.5404002624563873e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 704800, "loss": 0.5306462049484253, "lr": 8e-05, "grad_norm": 0.20430684528132304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103163135910, "tokens_per_second": 152653.186544431, "data_wait_s": 1.72220024978742e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 704900, "loss": 0.4992697238922119, "lr": 8e-05, "grad_norm": 0.2101155164146234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103177784453, "tokens_per_second": 152551.52564595608, "data_wait_s": 1.8695998733164743e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 705000, "loss": 0.5618807673454285, "lr": 8e-05, "grad_norm": 0.20961666563614817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103192342154, "tokens_per_second": 151723.58814000429, "data_wait_s": 1.3707998732570559e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c92e50681295f64b9bd6dbf016ca7b3ca982c19ef8b24b17e20674392f154729
-size 423996
+oid sha256:b7f3be4fc82ec001746fb02c3dc7b30e45bf4929ab905a6bbd7965d96c8f4c9b
+size 426114