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": 76072.48955060703,
-      "required_pages_per_second_upper_estimate": 0.006593770590284551
+      "required_token_positions_per_second": 76129.64169429647,
+      "required_pages_per_second_upper_estimate": 0.006598724393248768
     },
     {
       "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": 9509.061193825879,
-      "required_pages_per_second_upper_estimate": 0.003793306683351635
+      "required_token_positions_per_second": 9516.205211787059,
+      "required_pages_per_second_upper_estimate": 0.003796156538928937
     },
     {
       "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": 19018.122387651758,
-      "required_pages_per_second_upper_estimate": 0.004399507906860893
+      "required_token_positions_per_second": 19032.410423574118,
+      "required_pages_per_second_upper_estimate": 0.004402813192510687
     },
     {
       "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": 28527.183581477635,
-      "required_pages_per_second_upper_estimate": 0.0017371558095468868
+      "required_token_positions_per_second": 28548.615635361177,
+      "required_pages_per_second_upper_estimate": 0.001738460909183098
     },
     {
       "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": 28527.183581477635,
-      "required_pages_per_second_upper_estimate": 0.002196101560400498
+      "required_token_positions_per_second": 28548.615635361177,
+      "required_pages_per_second_upper_estimate": 0.002197751459236175
     },
     {
       "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": 28527.183581477635,
-      "required_pages_per_second_upper_estimate": 0.35421214574018944
+      "required_token_positions_per_second": 28548.615635361177,
+      "required_pages_per_second_upper_estimate": 0.3544782601209528
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.077365188468832,
-    "producer_p95_seconds": 0.20389741999861144,
-    "gpu_update_seconds": 0.9691808488921956,
-    "producer_batches_per_second": 12.925710126110124,
-    "gpu_batches_per_second": 1.0317991746772872,
-    "producer_headroom": 12.527350712557858,
+    "producer_batch_seconds": 0.21398998597658192,
+    "producer_p95_seconds": 0.2228557100006583,
+    "gpu_update_seconds": 0.9684532641840031,
+    "producer_batches_per_second": 4.673115872391502,
+    "gpu_batches_per_second": 1.032574350237311,
+    "producer_headroom": 4.525694320527626,
     "sustainable": true,
-    "buffer_seconds": 60.08921263131612,
+    "buffer_seconds": 60.044102379408194,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 558265.2179067562,
-    "upload_bytes_per_second": 5882472.776820311,
+    "generation_bytes_per_second": 521031.8566480682,
+    "upload_bytes_per_second": 5916053.49644013,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 119.79072001363768,
+    "estimated_full_shards_per_hour": 111.80130743581327,
     "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": [
@@ -448,6 +448,12 @@
       "bytes": 1197297,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 0.9917393740015541,
+      "bytes": 7505816,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -101,3 +101,4 @@
 {"step": 125100, "loss": 0.6330022923648357, "lr": 0.0003, "grad_norm": 0.15490170281847904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18307168581, "tokens_per_second": 131334.71537567364, "data_wait_s": 1.274699934583623e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 125200, "loss": 0.6066600568592548, "lr": 0.0003, "grad_norm": 0.14261032306092158, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18321829653, "tokens_per_second": 149559.8669171572, "data_wait_s": 1.3052000213065185e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 125300, "loss": 0.5879206787794828, "lr": 0.0003, "grad_norm": 0.1335140020427302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18336475000, "tokens_per_second": 149459.29210966578, "data_wait_s": 1.6451000192319043e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 125400, "loss": 0.6304312646389008, "lr": 0.0003, "grad_norm": 0.15239185870695368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18351272583, "tokens_per_second": 145620.47833194915, "data_wait_s": 9.3830003606854e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9f9a786a10413672c3f0628dc6af536d243c90053f38961495d7314607c5d2bc
-size 219885
+oid sha256:529230e173a9c0eba83837390656db371fa81e1886c9e3f424c5c041f6c47b0e
+size 222003