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": 75564.35753252536,
-      "required_pages_per_second_upper_estimate": 0.0065497269948060575
+      "required_token_positions_per_second": 75391.64955651609,
+      "required_pages_per_second_upper_estimate": 0.006534757105169937
     },
     {
       "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": 9445.54469156567,
-      "required_pages_per_second_upper_estimate": 0.003767969000943701
+      "required_token_positions_per_second": 9423.956194564511,
+      "required_pages_per_second_upper_estimate": 0.0037593570267131448
     },
     {
       "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": 18891.08938313134,
-      "required_pages_per_second_upper_estimate": 0.004370121056969615
+      "required_token_positions_per_second": 18847.912389129022,
+      "required_pages_per_second_upper_estimate": 0.00436013281929632
     },
     {
       "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": 28336.63407469701,
-      "required_pages_per_second_upper_estimate": 0.001725552344319945
+      "required_token_positions_per_second": 28271.868583693533,
+      "required_pages_per_second_upper_estimate": 0.0017216084657161036
     },
     {
       "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": 28336.63407469701,
-      "required_pages_per_second_upper_estimate": 0.0021814325318937305
+      "required_token_positions_per_second": 28271.868583693533,
+      "required_pages_per_second_upper_estimate": 0.0021764467051139293
     },
     {
       "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": 28336.63407469701,
-      "required_pages_per_second_upper_estimate": 0.35184615859414936
+      "required_token_positions_per_second": 28271.868583693533,
+      "required_pages_per_second_upper_estimate": 0.3510419879520411
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.09754178520299206,
-    "producer_p95_seconds": 0.26839475099768606,
-    "gpu_update_seconds": 0.9756980990444479,
-    "producer_batches_per_second": 10.252016588777025,
-    "gpu_batches_per_second": 1.0249071930952331,
-    "producer_headroom": 10.00287309704189,
+    "producer_batch_seconds": 0.1757417922734703,
+    "producer_p95_seconds": 0.32925739900019835,
+    "gpu_update_seconds": 0.9779332384116498,
+    "producer_batches_per_second": 5.690166164027214,
+    "gpu_batches_per_second": 1.0225646912504895,
+    "producer_headroom": 5.564602623887528,
     "sustainable": true,
-    "buffer_seconds": 60.493282140755774,
+    "buffer_seconds": 60.631860781522285,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 164,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 465772.54253073153,
-    "upload_bytes_per_second": 4937033.1359248785,
+    "generation_bytes_per_second": 569697.6231098112,
+    "upload_bytes_per_second": 4930761.821839304,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.9439449972292,
+    "estimated_full_shards_per_hour": 61.1219240187204,
     "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": 1.0455480339987844,
-      "bytes": 7415331,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.2304811680005514,
       "bytes": 10294757,
@@ -562,6 +556,12 @@
       "bytes": 1021485,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 0.9136627450025117,
+      "bytes": 6235523,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -261,3 +261,4 @@
 {"step": 381100, "loss": 0.6193846669048071, "lr": 0.0003, "grad_norm": 0.1722070226718004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55781639790, "tokens_per_second": 149361.8708315675, "data_wait_s": 2.0012997993035242e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 381200, "loss": 0.6016373839229345, "lr": 0.0003, "grad_norm": 0.16683431062133336, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55796295027, "tokens_per_second": 149648.2933052749, "data_wait_s": 1.2978998711332679e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 381300, "loss": 0.5606618449091911, "lr": 0.0003, "grad_norm": 0.14681621864688338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55811042513, "tokens_per_second": 150601.81990592802, "data_wait_s": 1.3439999747788534e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 381400, "loss": 0.5771404895931482, "lr": 0.0003, "grad_norm": 0.14902781861344497, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55825602854, "tokens_per_second": 148512.8178592142, "data_wait_s": 9.18500154512003e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fd3e817a5d7cdf363c5393a0d8a87d817b5fcad178d8ee6f8f78c65dd94145aa
-size 561114
+oid sha256:25d67b9597352f4da21086e1c3a411b48e0b4ef3632e46fd7e55dd3f539ffe82
+size 563232