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": 75400.63557047799,
-      "required_pages_per_second_upper_estimate": 0.00653553599008531
+      "required_token_positions_per_second": 75462.21096246573,
+      "required_pages_per_second_upper_estimate": 0.006540873189001404
     },
     {
       "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": 9425.079446309748,
-      "required_pages_per_second_upper_estimate": 0.0037598051086284298
+      "required_token_positions_per_second": 9432.776370308216,
+      "required_pages_per_second_upper_estimate": 0.0037628755266906878
     },
     {
       "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": 18850.158892619496,
-      "required_pages_per_second_upper_estimate": 0.00436065250834173
+      "required_token_positions_per_second": 18865.552740616433,
+      "required_pages_per_second_upper_estimate": 0.004364213604153349
     },
     {
       "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": 28275.238338929244,
-      "required_pages_per_second_upper_estimate": 0.0017218136661302191
+      "required_token_positions_per_second": 28298.32911092465,
+      "required_pages_per_second_upper_estimate": 0.0017232197729968213
     },
     {
       "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": 28275.238338929244,
-      "required_pages_per_second_upper_estimate": 0.002176706118200054
+      "required_token_positions_per_second": 28298.32911092465,
+      "required_pages_per_second_upper_estimate": 0.002178483709747608
     },
     {
       "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": 28275.238338929244,
-      "required_pages_per_second_upper_estimate": 0.3510838290342233
+      "required_token_positions_per_second": 28298.32911092465,
+      "required_pages_per_second_upper_estimate": 0.35137053914256366
     }
   ],
   "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": 0.11867581032032604,
-    "producer_p95_seconds": 0.38192668199917534,
-    "gpu_update_seconds": 0.9778166913604521,
-    "producer_batches_per_second": 8.426317016928987,
-    "gpu_batches_per_second": 1.022686571865207,
-    "producer_headroom": 8.239393425847776,
+    "producer_batch_seconds": 0.4678662598205392,
+    "producer_p95_seconds": 0.4028179220040329,
+    "gpu_update_seconds": 0.9770188159033889,
+    "producer_batches_per_second": 2.1373629301321553,
+    "gpu_batches_per_second": 1.0235217415699018,
+    "producer_headroom": 2.088243799153516,
     "sustainable": true,
-    "buffer_seconds": 60.62463486434803,
+    "buffer_seconds": 60.57516658601011,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 527835.351332048,
-    "upload_bytes_per_second": 5757549.27819952,
+    "generation_bytes_per_second": 452049.0512491977,
+    "upload_bytes_per_second": 5812289.791627157,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.26117901774482,
+    "estimated_full_shards_per_hour": 96.9992032347388,
     "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": 1.024480873002176,
-      "bytes": 7386490,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.0989981620004983,
       "bytes": 6127153,
@@ -562,6 +556,12 @@
       "bytes": 973266,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.851456552001764,
+      "bytes": 16129409,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -345,3 +345,4 @@
 {"step": 229500, "loss": 0.5922210942953825, "lr": 0.0003, "grad_norm": 0.12819053892824425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33592338050, "tokens_per_second": 149916.34407644212, "data_wait_s": 1.1964999430347234e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 229600, "loss": 0.6117744147777557, "lr": 0.0003, "grad_norm": 0.17466701664698814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33606929569, "tokens_per_second": 148710.77335447943, "data_wait_s": 1.6170000890269876e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 229700, "loss": 0.6110566407442093, "lr": 0.0003, "grad_norm": 0.15732614767708517, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33621456149, "tokens_per_second": 147668.8134239256, "data_wait_s": 1.0698000551201403e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 229800, "loss": 0.5975795947015285, "lr": 0.0003, "grad_norm": 0.14342664525978513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33636017226, "tokens_per_second": 147616.1027512308, "data_wait_s": 8.350005373358727e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:667386d2bff3a5d8d45a30e532d34c8b4d04ed0c16210b7d4f4909afcf22e91d
-size 739809
+oid sha256:b8ce3815ef80262c8bba5f40c4a8104afbd5eb35f866956ea42cec1ec6906684
+size 741927