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": 74629.83999908494,
-      "required_pages_per_second_upper_estimate": 0.0064687253835204765
+      "required_token_positions_per_second": 76056.42731777119,
+      "required_pages_per_second_upper_estimate": 0.006592378356651697
     },
     {
       "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": 9328.729999885618,
-      "required_pages_per_second_upper_estimate": 0.0037213698738972466
+      "required_token_positions_per_second": 9507.053414721398,
+      "required_pages_per_second_upper_estimate": 0.003792505750247885
     },
     {
       "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": 18657.459999771236,
-      "required_pages_per_second_upper_estimate": 0.004316075010865969
+      "required_token_positions_per_second": 19014.106829442797,
+      "required_pages_per_second_upper_estimate": 0.004398578978140661
     },
     {
       "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": 27986.189999656854,
-      "required_pages_per_second_upper_estimate": 0.0017042121387879638
+      "required_token_positions_per_second": 28521.160244164195,
+      "required_pages_per_second_upper_estimate": 0.0017367890199065069
     },
     {
       "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": 27986.189999656854,
-      "required_pages_per_second_upper_estimate": 0.0021544543769058512
+      "required_token_positions_per_second": 28521.160244164195,
+      "required_pages_per_second_upper_estimate": 0.0021956378672204446
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27986.189999656854,
-      "required_pages_per_second_upper_estimate": 0.3474948160430219
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28521.160244164195,
+      "required_pages_per_second_upper_estimate": 0.3541373560495697
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10398381299995663,
-    "producer_p95_seconds": 0.34360158699928434,
-    "gpu_update_seconds": 0.987915825638967,
-    "producer_batches_per_second": 9.61688142750081,
-    "gpu_batches_per_second": 1.012231987834811,
-    "producer_headroom": 9.500669355521508,
+    "producer_batch_seconds": 0.19188075915621994,
+    "producer_p95_seconds": 0.3397695979983837,
+    "gpu_update_seconds": 0.9693855286149218,
+    "producer_batches_per_second": 5.211569958329427,
+    "gpu_batches_per_second": 1.0315813167015406,
+    "producer_headroom": 5.052020498968818,
     "sustainable": true,
-    "buffer_seconds": 60.262865363976985,
+    "buffer_seconds": 60.10190277412515,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 498503.4143872507,
-    "upload_bytes_per_second": 4551850.697961663,
+    "generation_bytes_per_second": 489184.4316606625,
+    "upload_bytes_per_second": 4580002.025090962,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.96722816193716,
+    "estimated_full_shards_per_hour": 104.9675914036265,
     "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": 1.0286378179989697,
-      "bytes": 1051160,
-      "pages": 32,
-      "shards": 1
-    },
     {
       "seconds": 1.0477113340020878,
       "bytes": 9989334,
@@ -562,6 +556,12 @@
       "bytes": 2660846,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.5187353149995033,
+      "bytes": 5879969,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -316,3 +316,4 @@
 {"step": 186600, "loss": 0.6013043075799942, "lr": 0.0003, "grad_norm": 0.1489599878412404, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27310309660, "tokens_per_second": 150333.442953161, "data_wait_s": 6.770998879801482e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 186700, "loss": 0.56705092638731, "lr": 0.0003, "grad_norm": 0.14055031140339486, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27324912598, "tokens_per_second": 149501.57454912472, "data_wait_s": 1.713199890218675e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 186800, "loss": 0.633107615634799, "lr": 0.0003, "grad_norm": 0.14072124692690208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27339416525, "tokens_per_second": 148882.1327731205, "data_wait_s": 1.079800131265074e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 186900, "loss": 0.587534736841917, "lr": 0.0003, "grad_norm": 0.14738928761525955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27353807895, "tokens_per_second": 145889.4209428797, "data_wait_s": 1.0378000297350809e-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 sha256:53b4777196b9cbafac54c32a0e821a98422607304ad8e291c7aa5ab8806ef5f1
-size 678387
+oid sha256:66df3e349e889f03930d744b569df559ede36438532a8115985e18a7cdfe4da5
+size 680505