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": 75374.34807361432,
-      "required_pages_per_second_upper_estimate": 0.00653325745648222
+      "required_token_positions_per_second": 75599.36850148675,
+      "required_pages_per_second_upper_estimate": 0.006552761656861142
     },
     {
       "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": 9421.79350920179,
-      "required_pages_per_second_upper_estimate": 0.0037584942991869276
+      "required_token_positions_per_second": 9449.921062685844,
+      "required_pages_per_second_upper_estimate": 0.003769714800816118
     },
     {
       "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": 18843.58701840358,
-      "required_pages_per_second_upper_estimate": 0.004359132220902914
+      "required_token_positions_per_second": 18899.842125371688,
+      "required_pages_per_second_upper_estimate": 0.0043721458498439175
     },
     {
       "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": 28265.380527605375,
-      "required_pages_per_second_upper_estimate": 0.0017212133771405344
+      "required_token_positions_per_second": 28349.763188057528,
+      "required_pages_per_second_upper_estimate": 0.0017263518384404665
     },
     {
       "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": 28265.380527605375,
-      "required_pages_per_second_upper_estimate": 0.002175947236596174
+      "required_token_positions_per_second": 28349.763188057528,
+      "required_pages_per_second_upper_estimate": 0.0021824432473839354
     },
     {
       "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": 28265.380527605375,
-      "required_pages_per_second_upper_estimate": 0.35096142800955304
+      "required_token_positions_per_second": 28349.763188057528,
+      "required_pages_per_second_upper_estimate": 0.3520091782417712
     }
   ],
   "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": 0.12434348291424158,
-    "producer_p95_seconds": 0.33146621999912895,
-    "gpu_update_seconds": 0.9781577139213672,
-    "producer_batches_per_second": 8.042238938165257,
-    "gpu_batches_per_second": 1.022330024869986,
-    "producer_headroom": 7.866578054565132,
+    "producer_batch_seconds": 0.10500323269531009,
+    "producer_p95_seconds": 0.34588638599961996,
+    "gpu_update_seconds": 0.9752462416210534,
+    "producer_batches_per_second": 9.523516317842512,
+    "gpu_batches_per_second": 1.0253820597532382,
+    "producer_headroom": 9.287773495992683,
     "sustainable": true,
-    "buffer_seconds": 60.64577826312477,
+    "buffer_seconds": 60.46526698050531,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 304875.6133312581,
-    "upload_bytes_per_second": 5942677.3499470875,
+    "generation_bytes_per_second": 305888.0392322154,
+    "upload_bytes_per_second": 5978029.755433097,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.41920947984035,
+    "estimated_full_shards_per_hour": 65.63645251011701,
     "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": 1.146249319997878,
-      "bytes": 4617912,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 0.8623779709996597,
       "bytes": 6286843,
@@ -562,6 +556,12 @@
       "bytes": 6867540,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.0004031040007249,
+      "bytes": 6252534,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -279,3 +279,4 @@
 {"step": 112900, "loss": 0.672998858615756, "lr": 0.0003, "grad_norm": 0.15856373016345113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16520758339, "tokens_per_second": 149601.41419452222, "data_wait_s": 9.977004083339125e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 113000, "loss": 0.6347733959555626, "lr": 0.0003, "grad_norm": 0.14740347821185226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16535376761, "tokens_per_second": 149178.66304026908, "data_wait_s": 1.59679984790273e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 113100, "loss": 0.6114037893712521, "lr": 0.0003, "grad_norm": 0.1358469776640938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16550045050, "tokens_per_second": 149195.7667716262, "data_wait_s": 1.4092001947574317e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 113200, "loss": 0.5982640460133553, "lr": 0.0003, "grad_norm": 0.13941891006974785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16564794409, "tokens_per_second": 149804.71258791542, "data_wait_s": 1.2853997759521008e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7fa42e669d633c8b1c0b1ffded377f3dc1d7eaf821c4fca25e719e0ecf6dafab
-size 599238
+oid sha256:dbe4570a5209209c34e14cba7a4c932e4699585b8eaf72b7b762c578c639c2bf
+size 601356