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": 76066.89583139225,
-      "required_pages_per_second_upper_estimate": 0.006593285740880165
+      "required_token_positions_per_second": 75580.61455858574,
+      "required_pages_per_second_upper_estimate": 0.006551136112621916
     },
     {
       "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": 9508.361978924031,
-      "required_pages_per_second_upper_estimate": 0.0037930277560730935
+      "required_token_positions_per_second": 9447.576819823218,
+      "required_pages_per_second_upper_estimate": 0.0037687796472886622
     },
     {
       "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": 19016.723957848062,
-      "required_pages_per_second_upper_estimate": 0.004399184404737332
+      "required_token_positions_per_second": 18895.153639646436,
+      "required_pages_per_second_upper_estimate": 0.0043710612511859105
     },
     {
       "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": 28525.08593677209,
-      "required_pages_per_second_upper_estimate": 0.0017370280739898104
+      "required_token_positions_per_second": 28342.73045946965,
+      "required_pages_per_second_upper_estimate": 0.00172592358216734
     },
     {
       "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": 28525.08593677209,
-      "required_pages_per_second_upper_estimate": 0.002195940077904413
+      "required_token_positions_per_second": 28342.73045946965,
+      "required_pages_per_second_upper_estimate": 0.002181901848469402
     },
     {
       "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": 28525.08593677209,
-      "required_pages_per_second_upper_estimate": 0.3541861000133118
+      "required_token_positions_per_second": 28342.73045946965,
+      "required_pages_per_second_upper_estimate": 0.35192185528973824
     }
   ],
   "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": 0.12437562977314087,
+    "producer_batch_seconds": 0.21340095372633527,
     "producer_p95_seconds": 0.28320598299978883,
-    "gpu_update_seconds": 0.969252119390062,
-    "producer_batches_per_second": 8.040160293652251,
-    "gpu_batches_per_second": 1.0317233050047776,
-    "producer_headroom": 7.792942404858267,
+    "gpu_update_seconds": 0.9754882310840474,
+    "producer_batches_per_second": 4.686014671154642,
+    "gpu_batches_per_second": 1.025127693123179,
+    "producer_headroom": 4.571152162398536,
     "sustainable": true,
-    "buffer_seconds": 60.09363140218384,
+    "buffer_seconds": 60.48027032721094,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 457439.12992277864,
-    "upload_bytes_per_second": 5392405.612329192,
+    "generation_bytes_per_second": 431620.1992488283,
+    "upload_bytes_per_second": 5422944.361081789,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.15578864347953,
+    "estimated_full_shards_per_hour": 92.61564715479504,
     "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": 1.6021141640012502,
-      "bytes": 6045584,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.4088935970012244,
       "bytes": 4960887,
@@ -562,6 +556,12 @@
       "bytes": 2622896,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1337759699999879,
+      "bytes": 5871486,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -218,3 +218,4 @@
 {"step": 176800, "loss": 0.5874834768474102, "lr": 0.0003, "grad_norm": 0.132651790254468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25875330860, "tokens_per_second": 149868.69283462793, "data_wait_s": 1.1986998288193718e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 176900, "loss": 0.5908063501119614, "lr": 0.0003, "grad_norm": 0.1466829626245251, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25890070368, "tokens_per_second": 150322.62844061662, "data_wait_s": 1.2328000593697652e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 177000, "loss": 0.6187183447182178, "lr": 0.0003, "grad_norm": 0.13996301249884616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25904777042, "tokens_per_second": 149623.20916157932, "data_wait_s": 9.587001841282472e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 177100, "loss": 0.6259839870035648, "lr": 0.0003, "grad_norm": 0.14912587300424573, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25919474028, "tokens_per_second": 150611.38927202567, "data_wait_s": 1.5474997780984268e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eedb2ef57d7fc58e62c0b905768e7e7a6998b35bd5f9c6e4f404691f054d8432
-size 469257
+oid sha256:d3009b344952d4ffb0b019c681efb4c2bec347a1f43ac856dca7aad0f073cbe8
+size 471375