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": 75221.08669729417,
-      "required_pages_per_second_upper_estimate": 0.006519973148820196
+      "required_token_positions_per_second": 74921.70065387659,
+      "required_pages_per_second_upper_estimate": 0.00649402312536375
     },
     {
       "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": 9402.635837161772,
-      "required_pages_per_second_upper_estimate": 0.003750852017377442
+      "required_token_positions_per_second": 9365.212581734573,
+      "required_pages_per_second_upper_estimate": 0.003735923321260002
     },
     {
       "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": 18805.271674323543,
-      "required_pages_per_second_upper_estimate": 0.004350268640377021
+      "required_token_positions_per_second": 18730.425163469146,
+      "required_pages_per_second_upper_estimate": 0.004332954217344458
     },
     {
       "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": 28207.90751148531,
-      "required_pages_per_second_upper_estimate": 0.001717713571996434
+      "required_token_positions_per_second": 28095.63774520372,
+      "required_pages_per_second_upper_estimate": 0.0017108769322638257
     },
     {
       "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": 28207.90751148531,
-      "required_pages_per_second_upper_estimate": 0.0021715228047197604
+      "required_token_positions_per_second": 28095.63774520372,
+      "required_pages_per_second_upper_estimate": 0.0021628799673287995
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28207.90751148531,
-      "required_pages_per_second_upper_estimate": 0.3502478054991533
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28095.63774520372,
+      "required_pages_per_second_upper_estimate": 0.3488537907446729
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.0760199385312319,
-    "producer_p95_seconds": 0.17086536099986915,
-    "gpu_update_seconds": 0.9801506896157633,
-    "producer_batches_per_second": 13.154443680445253,
-    "gpu_batches_per_second": 1.0202512844142546,
-    "producer_headroom": 12.893337044900134,
+    "producer_batch_seconds": 0.17443918739062525,
+    "producer_p95_seconds": 0.17294827600017015,
+    "gpu_update_seconds": 0.9840673577420347,
+    "producer_batches_per_second": 5.732656835649432,
+    "gpu_batches_per_second": 1.0161906013166855,
+    "producer_headroom": 5.64132046509935,
     "sustainable": true,
-    "buffer_seconds": 60.76934275617733,
+    "buffer_seconds": 60.028108822264116,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 165,
+    "upload_pages_per_commit": 156,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 610145.5544581906,
-    "upload_bytes_per_second": 5602022.431596586,
+    "generation_bytes_per_second": 594509.0457440881,
+    "upload_bytes_per_second": 5755478.472061156,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.46151626257557,
+    "estimated_full_shards_per_hour": 63.783900877199095,
     "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": [
@@ -328,6 +328,12 @@
       "bytes": 7407679,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.307497912999679,
+      "bytes": 11505681,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -27,3 +27,4 @@
 {"step": 84600, "loss": 0.5803650245070457, "lr": 0.0003, "grad_norm": 0.12690611449244352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12380705618, "tokens_per_second": 149123.07797826442, "data_wait_s": 1.2340000012045493e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 84700, "loss": 0.6427894756197929, "lr": 0.0003, "grad_norm": 0.14702141787082534, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12395361653, "tokens_per_second": 147233.25360334205, "data_wait_s": 1.1217999599466566e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 84800, "loss": 0.6267599500715733, "lr": 0.0003, "grad_norm": 0.1399491145580503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12410053275, "tokens_per_second": 149362.03363209727, "data_wait_s": 1.4009000096848467e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 84900, "loss": 0.6152340620756149, "lr": 0.0003, "grad_norm": 0.12822516601325828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12424713818, "tokens_per_second": 147683.3784253226, "data_wait_s": 6.062000011297641e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cabf22cae536de552a9137bd764b476dd5f3f329e571c9bd3932ba408a317d95
-size 61587
+oid sha256:602fcaade8854c150362bbcaf31f714348305f0290b5944e4ba809482ae8f9b7
+size 63705