Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77215.19644552871,
-      "required_pages_per_second_upper_estimate": 0.005341926454041404
+      "required_token_positions_per_second": 77240.39749051181,
+      "required_pages_per_second_upper_estimate": 0.005343669920807815
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9651.899555691089,
-      "required_pages_per_second_upper_estimate": 0.0040410487893856005
+      "required_token_positions_per_second": 9655.049686313976,
+      "required_pages_per_second_upper_estimate": 0.0040423676832951954
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19303.799111382177,
-      "required_pages_per_second_upper_estimate": 0.004020472279484501
+      "required_token_positions_per_second": 19310.099372627952,
+      "required_pages_per_second_upper_estimate": 0.004021784457752934
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28955.698667073266,
-      "required_pages_per_second_upper_estimate": 0.0017358499222991813
+      "required_token_positions_per_second": 28965.149058941926,
+      "required_pages_per_second_upper_estimate": 0.0017364164588617947
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28955.698667073266,
-      "required_pages_per_second_upper_estimate": 0.0022511782087066105
+      "required_token_positions_per_second": 28965.149058941926,
+      "required_pages_per_second_upper_estimate": 0.0022519129351065183
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28955.698667073266,
-      "required_pages_per_second_upper_estimate": 0.3714126122942659
+      "required_token_positions_per_second": 28965.149058941926,
+      "required_pages_per_second_upper_estimate": 0.37153383177411686
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13976823710152075,
-    "producer_p95_seconds": 0.3075653330015484,
-    "gpu_update_seconds": 0.9548379515165937,
-    "producer_batches_per_second": 7.154701388081824,
-    "gpu_batches_per_second": 1.047298128872731,
-    "producer_headroom": 6.831580417108978,
+    "producer_batch_seconds": 0.07962050917953434,
+    "producer_p95_seconds": 0.280256891001045,
+    "gpu_update_seconds": 0.954526418757189,
+    "producer_batches_per_second": 12.559578057270702,
+    "gpu_batches_per_second": 1.0476399399212213,
+    "producer_headroom": 11.988449064107977,
     "sustainable": true,
-    "buffer_seconds": 60.1547909455454,
+    "buffer_seconds": 60.13516438170291,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315402.5068014548,
+    "generation_bytes_per_second": 315734.3997911154,
     "upload_bytes_per_second": 5142131.291503559,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.67803576500638,
+    "estimated_full_shards_per_hour": 67.74925227451416,
     "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": [

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -264,3 +264,4 @@
 {"step": 951400, "loss": 0.5310010313987732, "lr": 3e-05, "grad_norm": 0.3356313779004489, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139258878320, "tokens_per_second": 152646.37032217704, "data_wait_s": 1.300000076298602e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 951500, "loss": 0.5428860187530518, "lr": 3e-05, "grad_norm": 0.3392346283253021, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139273610388, "tokens_per_second": 153114.31166177063, "data_wait_s": 1.4474000636255369e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 951600, "loss": 0.562923014163971, "lr": 3e-05, "grad_norm": 0.3446790800626669, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139288352143, "tokens_per_second": 152546.54236023803, "data_wait_s": 1.176499790744856e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 951700, "loss": 0.500239372253418, "lr": 3e-05, "grad_norm": 0.34763622765609353, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139303135164, "tokens_per_second": 153730.24730144194, "data_wait_s": 1.4639001165051013e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ef0ccd02bebe0dfe6781e115903758c8298035e47c3f513c149d5ed5ba3d6314
-size 567468
+oid sha256:53242cfc026576cafc9e3110fd4eddb8d0ad94f8f8e7f22b9b27aa840b26a48f
+size 569586